This is page 5 of 5. Use http://codebase.md/rashidazarang/airtable-mcp?lines=true&page={x} to view the full context.
# Directory Structure
```
├── .eslintrc.js
├── .github
│ ├── ISSUE_TEMPLATE
│ │ ├── bug_report.md
│ │ ├── custom.md
│ │ └── feature_request.md
│ └── pull_request_template.md
├── .gitignore
├── .nvmrc
├── .prettierrc
├── bin
│ ├── airtable-crud-cli.js
│ └── airtable-mcp.js
├── CHANGELOG.md
├── CODE_OF_CONDUCT.md
├── CONTRIBUTING.md
├── docker
│ ├── Dockerfile
│ └── Dockerfile.node
├── docs
│ ├── guides
│ │ ├── CLAUDE_INTEGRATION.md
│ │ ├── ENHANCED_FEATURES.md
│ │ ├── INSTALLATION.md
│ │ └── QUICK_START.md
│ └── releases
│ ├── RELEASE_NOTES_v1.2.2.md
│ ├── RELEASE_NOTES_v1.2.4.md
│ ├── RELEASE_NOTES_v1.4.0.md
│ ├── RELEASE_NOTES_v1.5.0.md
│ └── RELEASE_NOTES_v1.6.0.md
├── examples
│ ├── airtable-crud-example.js
│ ├── building-mcp.md
│ ├── claude_config.json
│ ├── claude_simple_config.json
│ ├── env-demo.js
│ ├── example_usage.md
│ ├── example-tasks-update.json
│ ├── example-tasks.json
│ ├── python_debug_patch.txt
│ ├── sample-transform.js
│ ├── typescript
│ │ ├── advanced-ai-prompts.ts
│ │ ├── basic-usage.ts
│ │ └── claude-desktop-config.json
│ └── windsurf_mcp_config.json
├── index.js
├── ISSUE_RESPONSES.md
├── jest.config.js
├── LICENSE
├── package-lock.json
├── package.json
├── PROJECT_STRUCTURE.md
├── README.md
├── RELEASE_SUMMARY_v3.2.x.md
├── RELEASE_v3.2.1.md
├── RELEASE_v3.2.3.md
├── RELEASE_v3.2.4.md
├── requirements.txt
├── SECURITY_NOTICE.md
├── smithery.yaml
├── src
│ ├── index.js
│ ├── javascript
│ │ ├── airtable_simple_production.js
│ │ └── airtable_simple.js
│ ├── python
│ │ ├── airtable_mcp
│ │ │ ├── __init__.py
│ │ │ └── src
│ │ │ └── server.py
│ │ ├── inspector_server.py
│ │ ├── inspector.py
│ │ ├── setup.py
│ │ ├── simple_airtable_server.py
│ │ └── test_client.py
│ └── typescript
│ ├── ai-prompts.d.ts
│ ├── airtable-mcp-server.d.ts
│ ├── airtable-mcp-server.ts
│ ├── errors.ts
│ ├── index.d.ts
│ ├── prompt-templates.ts
│ ├── test-suite.d.ts
│ ├── test-suite.ts
│ ├── tools-schemas.ts
│ └── tools.d.ts
├── TESTING_REPORT.md
├── tests
│ ├── test_all_features.sh
│ ├── test_mcp_comprehensive.js
│ ├── test_v1.5.0_final.sh
│ └── test_v1.6.0_comprehensive.sh
├── tsconfig.json
└── types
└── typescript
├── airtable-mcp-server.d.ts
├── errors.d.ts
├── prompt-templates.d.ts
├── test-suite.d.ts
└── tools-schemas.d.ts
```
# Files
--------------------------------------------------------------------------------
/examples/building-mcp.md:
--------------------------------------------------------------------------------
```markdown
1 | # Example Clients
2 | Source: https://modelcontextprotocol.io/clients
3 |
4 | A list of applications that support MCP integrations
5 |
6 | This page provides an overview of applications that support the Model Context Protocol (MCP). Each client may support different MCP features, allowing for varying levels of integration with MCP servers.
7 |
8 | ## Feature support matrix
9 |
10 | | Client | [Resources] | [Prompts] | [Tools] | [Sampling] | Roots | Notes |
11 | | ------------------------------------ | ----------- | --------- | ------- | ---------- | ----- | ------------------------------------------------------------------ |
12 | | [Claude Desktop App][Claude] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
13 | | [5ire][5ire] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
14 | | [BeeAI Framework][BeeAI Framework] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in agentic workflows. |
15 | | [Cline][Cline] | ✅ | ❌ | ✅ | ❌ | ❌ | Supports tools and resources. |
16 | | [Continue][Continue] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
17 | | [Cursor][Cursor] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
18 | | [Emacs Mcp][Mcp.el] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in Emacs. |
19 | | [Firebase Genkit][Genkit] | ⚠️ | ✅ | ✅ | ❌ | ❌ | Supports resource list and lookup through tools. |
20 | | [GenAIScript][GenAIScript] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
21 | | [Goose][Goose] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
22 | | [LibreChat][LibreChat] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Agents |
23 | | [mcp-agent][mcp-agent] | ❌ | ❌ | ✅ | ⚠️ | ❌ | Supports tools, server connection management, and agent workflows. |
24 | | [oterm][oterm] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
25 | | [Roo Code][Roo Code] | ✅ | ❌ | ✅ | ❌ | ❌ | Supports tools and resources. |
26 | | [Sourcegraph Cody][Cody] | ✅ | ❌ | ❌ | ❌ | ❌ | Supports resources through OpenCTX |
27 | | [Superinterface][Superinterface] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools |
28 | | [TheiaAI/TheiaIDE][TheiaAI/TheiaIDE] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Agents in Theia AI and the AI-powered Theia IDE |
29 | | [Windsurf Editor][Windsurf] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools with AI Flow for collaborative development. |
30 | | [Zed][Zed] | ❌ | ✅ | ❌ | ❌ | ❌ | Prompts appear as slash commands |
31 | | [SpinAI][SpinAI] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Typescript AI Agents |
32 | | [OpenSumi][OpenSumi] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in OpenSumi |
33 | | [Daydreams Agents][Daydreams] | ✅ | ✅ | ✅ | ❌ | ❌ | Support for drop in Servers to Daydreams agents |
34 |
35 | [Claude]: https://claude.ai/download
36 |
37 | [Cursor]: https://cursor.com
38 |
39 | [Zed]: https://zed.dev
40 |
41 | [Cody]: https://sourcegraph.com/cody
42 |
43 | [Genkit]: https://github.com/firebase/genkit
44 |
45 | [Continue]: https://github.com/continuedev/continue
46 |
47 | [GenAIScript]: https://microsoft.github.io/genaiscript/reference/scripts/mcp-tools/
48 |
49 | [Cline]: https://github.com/cline/cline
50 |
51 | [LibreChat]: https://github.com/danny-avila/LibreChat
52 |
53 | [TheiaAI/TheiaIDE]: https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/
54 |
55 | [Superinterface]: https://superinterface.ai
56 |
57 | [5ire]: https://github.com/nanbingxyz/5ire
58 |
59 | [BeeAI Framework]: https://i-am-bee.github.io/beeai-framework
60 |
61 | [mcp-agent]: https://github.com/lastmile-ai/mcp-agent
62 |
63 | [Mcp.el]: https://github.com/lizqwerscott/mcp.el
64 |
65 | [Roo Code]: https://roocode.com
66 |
67 | [Goose]: https://block.github.io/goose/docs/goose-architecture/#interoperability-with-extensions
68 |
69 | [Windsurf]: https://codeium.com/windsurf
70 |
71 | [Daydreams]: https://github.com/daydreamsai/daydreams
72 |
73 | [SpinAI]: https://spinai.dev
74 |
75 | [OpenSumi]: https://github.com/opensumi/core
76 |
77 | [oterm]: https://github.com/ggozad/oterm
78 |
79 | [Resources]: https://modelcontextprotocol.io/docs/concepts/resources
80 |
81 | [Prompts]: https://modelcontextprotocol.io/docs/concepts/prompts
82 |
83 | [Tools]: https://modelcontextprotocol.io/docs/concepts/tools
84 |
85 | [Sampling]: https://modelcontextprotocol.io/docs/concepts/sampling
86 |
87 | ## Client details
88 |
89 | ### Claude Desktop App
90 |
91 | The Claude desktop application provides comprehensive support for MCP, enabling deep integration with local tools and data sources.
92 |
93 | **Key features:**
94 |
95 | * Full support for resources, allowing attachment of local files and data
96 | * Support for prompt templates
97 | * Tool integration for executing commands and scripts
98 | * Local server connections for enhanced privacy and security
99 |
100 | > ⓘ Note: The Claude.ai web application does not currently support MCP. MCP features are only available in the desktop application.
101 |
102 | ### 5ire
103 |
104 | [5ire](https://github.com/nanbingxyz/5ire) is an open source cross-platform desktop AI assistant that supports tools through MCP servers.
105 |
106 | **Key features:**
107 |
108 | * Built-in MCP servers can be quickly enabled and disabled.
109 | * Users can add more servers by modifying the configuration file.
110 | * It is open-source and user-friendly, suitable for beginners.
111 | * Future support for MCP will be continuously improved.
112 |
113 | ### BeeAI Framework
114 |
115 | [BeeAI Framework](https://i-am-bee.github.io/beeai-framework) is an open-source framework for building, deploying, and serving powerful agentic workflows at scale. The framework includes the **MCP Tool**, a native feature that simplifies the integration of MCP servers into agentic workflows.
116 |
117 | **Key features:**
118 |
119 | * Seamlessly incorporate MCP tools into agentic workflows.
120 | * Quickly instantiate framework-native tools from connected MCP client(s).
121 | * Planned future support for agentic MCP capabilities.
122 |
123 | **Learn more:**
124 |
125 | * [Example of using MCP tools in agentic workflow](https://i-am-bee.github.io/beeai-framework/#/typescript/tools?id=using-the-mcptool-class)
126 |
127 | ### Cline
128 |
129 | [Cline](https://github.com/cline/cline) is an autonomous coding agent in VS Code that edits files, runs commands, uses a browser, and more–with your permission at each step.
130 |
131 | **Key features:**
132 |
133 | * Create and add tools through natural language (e.g. "add a tool that searches the web")
134 | * Share custom MCP servers Cline creates with others via the `~/Documents/Cline/MCP` directory
135 | * Displays configured MCP servers along with their tools, resources, and any error logs
136 |
137 | ### Continue
138 |
139 | [Continue](https://github.com/continuedev/continue) is an open-source AI code assistant, with built-in support for all MCP features.
140 |
141 | **Key features**
142 |
143 | * Type "@" to mention MCP resources
144 | * Prompt templates surface as slash commands
145 | * Use both built-in and MCP tools directly in chat
146 | * Supports VS Code and JetBrains IDEs, with any LLM
147 |
148 | ### Cursor
149 |
150 | [Cursor](https://docs.cursor.com/advanced/model-context-protocol) is an AI code editor.
151 |
152 | **Key Features**:
153 |
154 | * Support for MCP tools in Cursor Composer
155 | * Support for both STDIO and SSE
156 |
157 | ### Emacs Mcp
158 |
159 | [Emacs Mcp](https://github.com/lizqwerscott/mcp.el) is an Emacs client designed to interface with MCP servers, enabling seamless connections and interactions. It provides MCP tool invocation support for AI plugins like [gptel](https://github.com/karthink/gptel) and [llm](https://github.com/ahyatt/llm), adhering to Emacs' standard tool invocation format. This integration enhances the functionality of AI tools within the Emacs ecosystem.
160 |
161 | **Key features:**
162 |
163 | * Provides MCP tool support for Emacs.
164 |
165 | ### Firebase Genkit
166 |
167 | [Genkit](https://github.com/firebase/genkit) is Firebase's SDK for building and integrating GenAI features into applications. The [genkitx-mcp](https://github.com/firebase/genkit/tree/main/js/plugins/mcp) plugin enables consuming MCP servers as a client or creating MCP servers from Genkit tools and prompts.
168 |
169 | **Key features:**
170 |
171 | * Client support for tools and prompts (resources partially supported)
172 | * Rich discovery with support in Genkit's Dev UI playground
173 | * Seamless interoperability with Genkit's existing tools and prompts
174 | * Works across a wide variety of GenAI models from top providers
175 |
176 | ### GenAIScript
177 |
178 | Programmatically assemble prompts for LLMs using [GenAIScript](https://microsoft.github.io/genaiscript/) (in JavaScript). Orchestrate LLMs, tools, and data in JavaScript.
179 |
180 | **Key features:**
181 |
182 | * JavaScript toolbox to work with prompts
183 | * Abstraction to make it easy and productive
184 | * Seamless Visual Studio Code integration
185 |
186 | ### Goose
187 |
188 | [Goose](https://github.com/block/goose) is an open source AI agent that supercharges your software development by automating coding tasks.
189 |
190 | **Key features:**
191 |
192 | * Expose MCP functionality to Goose through tools.
193 | * MCPs can be installed directly via the [extensions directory](https://block.github.io/goose/v1/extensions/), CLI, or UI.
194 | * Goose allows you to extend its functionality by [building your own MCP servers](https://block.github.io/goose/docs/tutorials/custom-extensions).
195 | * Includes built-in tools for development, web scraping, automation, memory, and integrations with JetBrains and Google Drive.
196 |
197 | ### LibreChat
198 |
199 | [LibreChat](https://github.com/danny-avila/LibreChat) is an open-source, customizable AI chat UI that supports multiple AI providers, now including MCP integration.
200 |
201 | **Key features:**
202 |
203 | * Extend current tool ecosystem, including [Code Interpreter](https://www.librechat.ai/docs/features/code_interpreter) and Image generation tools, through MCP servers
204 | * Add tools to customizable [Agents](https://www.librechat.ai/docs/features/agents), using a variety of LLMs from top providers
205 | * Open-source and self-hostable, with secure multi-user support
206 | * Future roadmap includes expanded MCP feature support
207 |
208 | ### mcp-agent
209 |
210 | [mcp-agent] is a simple, composable framework to build agents using Model Context Protocol.
211 |
212 | **Key features:**
213 |
214 | * Automatic connection management of MCP servers.
215 | * Expose tools from multiple servers to an LLM.
216 | * Implements every pattern defined in [Building Effective Agents](https://www.anthropic.com/research/building-effective-agents).
217 | * Supports workflow pause/resume signals, such as waiting for human feedback.
218 |
219 | ### oterm
220 |
221 | [oterm] is a terminal client for Ollama allowing users to create chats/agents.
222 |
223 | **Key features:**
224 |
225 | * Support for multiple fully customizable chat sessions with Ollama connected with tools.
226 | * Support for MCP tools.
227 |
228 | ### Roo Code
229 |
230 | [Roo Code](https://roocode.com) enables AI coding assistance via MCP.
231 |
232 | **Key features:**
233 |
234 | * Support for MCP tools and resources
235 | * Integration with development workflows
236 | * Extensible AI capabilities
237 |
238 | ### Sourcegraph Cody
239 |
240 | [Cody](https://openctx.org/docs/providers/modelcontextprotocol) is Sourcegraph's AI coding assistant, which implements MCP through OpenCTX.
241 |
242 | **Key features:**
243 |
244 | * Support for MCP resources
245 | * Integration with Sourcegraph's code intelligence
246 | * Uses OpenCTX as an abstraction layer
247 | * Future support planned for additional MCP features
248 |
249 | ### SpinAI
250 |
251 | [SpinAI](https://spinai.dev) is an open-source TypeScript framework for building observable AI agents. The framework provides native MCP compatibility, allowing agents to seamlessly integrate with MCP servers and tools.
252 |
253 | **Key features:**
254 |
255 | * Built-in MCP compatibility for AI agents
256 | * Open-source TypeScript framework
257 | * Observable agent architecture
258 | * Native support for MCP tools integration
259 |
260 | ### Superinterface
261 |
262 | [Superinterface](https://superinterface.ai) is AI infrastructure and a developer platform to build in-app AI assistants with support for MCP, interactive components, client-side function calling and more.
263 |
264 | **Key features:**
265 |
266 | * Use tools from MCP servers in assistants embedded via React components or script tags
267 | * SSE transport support
268 | * Use any AI model from any AI provider (OpenAI, Anthropic, Ollama, others)
269 |
270 | ### TheiaAI/TheiaIDE
271 |
272 | [Theia AI](https://eclipsesource.com/blogs/2024/10/07/introducing-theia-ai/) is a framework for building AI-enhanced tools and IDEs. The [AI-powered Theia IDE](https://eclipsesource.com/blogs/2024/10/08/introducting-ai-theia-ide/) is an open and flexible development environment built on Theia AI.
273 |
274 | **Key features:**
275 |
276 | * **Tool Integration**: Theia AI enables AI agents, including those in the Theia IDE, to utilize MCP servers for seamless tool interaction.
277 | * **Customizable Prompts**: The Theia IDE allows users to define and adapt prompts, dynamically integrating MCP servers for tailored workflows.
278 | * **Custom agents**: The Theia IDE supports creating custom agents that leverage MCP capabilities, enabling users to design dedicated workflows on the fly.
279 |
280 | Theia AI and Theia IDE's MCP integration provide users with flexibility, making them powerful platforms for exploring and adapting MCP.
281 |
282 | **Learn more:**
283 |
284 | * [Theia IDE and Theia AI MCP Announcement](https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/)
285 | * [Download the AI-powered Theia IDE](https://theia-ide.org/)
286 |
287 | ### Windsurf Editor
288 |
289 | [Windsurf Editor](https://codeium.com/windsurf) is an agentic IDE that combines AI assistance with developer workflows. It features an innovative AI Flow system that enables both collaborative and independent AI interactions while maintaining developer control.
290 |
291 | **Key features:**
292 |
293 | * Revolutionary AI Flow paradigm for human-AI collaboration
294 | * Intelligent code generation and understanding
295 | * Rich development tools with multi-model support
296 |
297 | ### Zed
298 |
299 | [Zed](https://zed.dev/docs/assistant/model-context-protocol) is a high-performance code editor with built-in MCP support, focusing on prompt templates and tool integration.
300 |
301 | **Key features:**
302 |
303 | * Prompt templates surface as slash commands in the editor
304 | * Tool integration for enhanced coding workflows
305 | * Tight integration with editor features and workspace context
306 | * Does not support MCP resources
307 |
308 | ### OpenSumi
309 |
310 | [OpenSumi](https://github.com/opensumi/core) is a framework helps you quickly build AI Native IDE products.
311 |
312 | **Key features:**
313 |
314 | * Supports MCP tools in OpenSumi
315 | * Supports built-in IDE MCP servers and custom MCP servers
316 |
317 | ### Daydreams
318 |
319 | [Daydreams](https://github.com/daydreamsai/daydreams) is a generative agent framework for executing anything onchain
320 |
321 | **Key features:**
322 |
323 | * Supports MCP Servers in config
324 | * Exposes MCP Client
325 |
326 | ## Adding MCP support to your application
327 |
328 | If you've added MCP support to your application, we encourage you to submit a pull request to add it to this list. MCP integration can provide your users with powerful contextual AI capabilities and make your application part of the growing MCP ecosystem.
329 |
330 | Benefits of adding MCP support:
331 |
332 | * Enable users to bring their own context and tools
333 | * Join a growing ecosystem of interoperable AI applications
334 | * Provide users with flexible integration options
335 | * Support local-first AI workflows
336 |
337 | To get started with implementing MCP in your application, check out our [Python](https://github.com/modelcontextprotocol/python-sdk) or [TypeScript SDK Documentation](https://github.com/modelcontextprotocol/typescript-sdk)
338 |
339 | ## Updates and corrections
340 |
341 | This list is maintained by the community. If you notice any inaccuracies or would like to update information about MCP support in your application, please submit a pull request or [open an issue in our documentation repository](https://github.com/modelcontextprotocol/docs/issues).
342 |
343 |
344 | # Contributing
345 | Source: https://modelcontextprotocol.io/development/contributing
346 |
347 | How to participate in Model Context Protocol development
348 |
349 | We welcome contributions from the community! Please review our [contributing guidelines](https://github.com/modelcontextprotocol/.github/blob/main/CONTRIBUTING.md) for details on how to submit changes.
350 |
351 | All contributors must adhere to our [Code of Conduct](https://github.com/modelcontextprotocol/.github/blob/main/CODE_OF_CONDUCT.md).
352 |
353 | For questions and discussions, please use [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions).
354 |
355 |
356 | # Roadmap
357 | Source: https://modelcontextprotocol.io/development/roadmap
358 |
359 | Our plans for evolving Model Context Protocol (H1 2025)
360 |
361 | The Model Context Protocol is rapidly evolving. This page outlines our current thinking on key priorities and future direction for **the first half of 2025**, though these may change significantly as the project develops.
362 |
363 | <Note>The ideas presented here are not commitments—we may solve these challenges differently than described, or some may not materialize at all. This is also not an *exhaustive* list; we may incorporate work that isn't mentioned here.</Note>
364 |
365 | We encourage community participation! Each section links to relevant discussions where you can learn more and contribute your thoughts.
366 |
367 | ## Remote MCP Support
368 |
369 | Our top priority is improving [remote MCP connections](https://github.com/modelcontextprotocol/specification/discussions/112), allowing clients to securely connect to MCP servers over the internet. Key initiatives include:
370 |
371 | * [**Authentication & Authorization**](https://github.com/modelcontextprotocol/specification/discussions/64): Adding standardized auth capabilities, particularly focused on OAuth 2.0 support.
372 |
373 | * [**Service Discovery**](https://github.com/modelcontextprotocol/specification/discussions/69): Defining how clients can discover and connect to remote MCP servers.
374 |
375 | * [**Stateless Operations**](https://github.com/modelcontextprotocol/specification/discussions/102): Thinking about whether MCP could encompass serverless environments too, where they will need to be mostly stateless.
376 |
377 | ## Reference Implementations
378 |
379 | To help developers build with MCP, we want to offer documentation for:
380 |
381 | * **Client Examples**: Comprehensive reference client implementation(s), demonstrating all protocol features
382 | * **Protocol Drafting**: Streamlined process for proposing and incorporating new protocol features
383 |
384 | ## Distribution & Discovery
385 |
386 | Looking ahead, we're exploring ways to make MCP servers more accessible. Some areas we may investigate include:
387 |
388 | * **Package Management**: Standardized packaging format for MCP servers
389 | * **Installation Tools**: Simplified server installation across MCP clients
390 | * **Sandboxing**: Improved security through server isolation
391 | * **Server Registry**: A common directory for discovering available MCP servers
392 |
393 | ## Agent Support
394 |
395 | We're expanding MCP's capabilities for [complex agentic workflows](https://github.com/modelcontextprotocol/specification/discussions/111), particularly focusing on:
396 |
397 | * [**Hierarchical Agent Systems**](https://github.com/modelcontextprotocol/specification/discussions/94): Improved support for trees of agents through namespacing and topology awareness.
398 |
399 | * [**Interactive Workflows**](https://github.com/modelcontextprotocol/specification/issues/97): Better handling of user permissions and information requests across agent hierarchies, and ways to send output to users instead of models.
400 |
401 | * [**Streaming Results**](https://github.com/modelcontextprotocol/specification/issues/117): Real-time updates from long-running agent operations.
402 |
403 | ## Broader Ecosystem
404 |
405 | We're also invested in:
406 |
407 | * **Community-Led Standards Development**: Fostering a collaborative ecosystem where all AI providers can help shape MCP as an open standard through equal participation and shared governance, ensuring it meets the needs of diverse AI applications and use cases.
408 | * [**Additional Modalities**](https://github.com/modelcontextprotocol/specification/discussions/88): Expanding beyond text to support audio, video, and other formats.
409 | * \[**Standardization**] Considering standardization through a standardization body.
410 |
411 | ## Get Involved
412 |
413 | We welcome community participation in shaping MCP's future. Visit our [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions) to join the conversation and contribute your ideas.
414 |
415 |
416 | # What's New
417 | Source: https://modelcontextprotocol.io/development/updates
418 |
419 | The latest updates and improvements to MCP
420 |
421 | <Update label="2025-02-14" description="Java SDK released">
422 | * We're excited to announce that the Java SDK developed by Spring AI at VMware Tanzu is now
423 | the official [Java SDK](https://github.com/modelcontextprotocol/java-sdk) for MCP.
424 | This joins our existing Kotlin SDK in our growing list of supported languages.
425 | The Spring AI team will maintain the SDK as an integral part of the Model Context Protocol
426 | organization. We're thrilled to welcome them to the MCP community!
427 | </Update>
428 |
429 | <Update label="2025-01-27" description="Python SDK 1.2.1">
430 | * Version [1.2.1](https://github.com/modelcontextprotocol/python-sdk/releases/tag/v1.2.1) of the MCP Python SDK has been released,
431 | delivering important stability improvements and bug fixes.
432 | </Update>
433 |
434 | <Update label="2025-01-18" description="SDK and Server Improvements">
435 | * Simplified, express-like API in the [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)
436 | * Added 8 new clients to the [clients page](https://modelcontextprotocol.io/clients)
437 | </Update>
438 |
439 | <Update label="2025-01-03" description="SDK and Server Improvements">
440 | * FastMCP API in the [Python SDK](https://github.com/modelcontextprotocol/python-sdk)
441 | * Dockerized MCP servers in the [servers repo](https://github.com/modelcontextprotocol/servers)
442 | </Update>
443 |
444 | <Update label="2024-12-21" description="Kotlin SDK released">
445 | * Jetbrains released a Kotlin SDK for MCP!
446 | * For a sample MCP Kotlin server, check out [this repository](https://github.com/modelcontextprotocol/kotlin-sdk/tree/main/samples/kotlin-mcp-server)
447 | </Update>
448 |
449 |
450 | # Core architecture
451 | Source: https://modelcontextprotocol.io/docs/concepts/architecture
452 |
453 | Understand how MCP connects clients, servers, and LLMs
454 |
455 | The Model Context Protocol (MCP) is built on a flexible, extensible architecture that enables seamless communication between LLM applications and integrations. This document covers the core architectural components and concepts.
456 |
457 | ## Overview
458 |
459 | MCP follows a client-server architecture where:
460 |
461 | * **Hosts** are LLM applications (like Claude Desktop or IDEs) that initiate connections
462 | * **Clients** maintain 1:1 connections with servers, inside the host application
463 | * **Servers** provide context, tools, and prompts to clients
464 |
465 | ```mermaid
466 | flowchart LR
467 | subgraph "Host"
468 | client1[MCP Client]
469 | client2[MCP Client]
470 | end
471 | subgraph "Server Process"
472 | server1[MCP Server]
473 | end
474 | subgraph "Server Process"
475 | server2[MCP Server]
476 | end
477 |
478 | client1 <-->|Transport Layer| server1
479 | client2 <-->|Transport Layer| server2
480 | ```
481 |
482 | ## Core components
483 |
484 | ### Protocol layer
485 |
486 | The protocol layer handles message framing, request/response linking, and high-level communication patterns.
487 |
488 | <Tabs>
489 | <Tab title="TypeScript">
490 | ```typescript
491 | class Protocol<Request, Notification, Result> {
492 | // Handle incoming requests
493 | setRequestHandler<T>(schema: T, handler: (request: T, extra: RequestHandlerExtra) => Promise<Result>): void
494 |
495 | // Handle incoming notifications
496 | setNotificationHandler<T>(schema: T, handler: (notification: T) => Promise<void>): void
497 |
498 | // Send requests and await responses
499 | request<T>(request: Request, schema: T, options?: RequestOptions): Promise<T>
500 |
501 | // Send one-way notifications
502 | notification(notification: Notification): Promise<void>
503 | }
504 | ```
505 | </Tab>
506 |
507 | <Tab title="Python">
508 | ```python
509 | class Session(BaseSession[RequestT, NotificationT, ResultT]):
510 | async def send_request(
511 | self,
512 | request: RequestT,
513 | result_type: type[Result]
514 | ) -> Result:
515 | """
516 | Send request and wait for response. Raises McpError if response contains error.
517 | """
518 | # Request handling implementation
519 |
520 | async def send_notification(
521 | self,
522 | notification: NotificationT
523 | ) -> None:
524 | """Send one-way notification that doesn't expect response."""
525 | # Notification handling implementation
526 |
527 | async def _received_request(
528 | self,
529 | responder: RequestResponder[ReceiveRequestT, ResultT]
530 | ) -> None:
531 | """Handle incoming request from other side."""
532 | # Request handling implementation
533 |
534 | async def _received_notification(
535 | self,
536 | notification: ReceiveNotificationT
537 | ) -> None:
538 | """Handle incoming notification from other side."""
539 | # Notification handling implementation
540 | ```
541 | </Tab>
542 | </Tabs>
543 |
544 | Key classes include:
545 |
546 | * `Protocol`
547 | * `Client`
548 | * `Server`
549 |
550 | ### Transport layer
551 |
552 | The transport layer handles the actual communication between clients and servers. MCP supports multiple transport mechanisms:
553 |
554 | 1. **Stdio transport**
555 | * Uses standard input/output for communication
556 | * Ideal for local processes
557 |
558 | 2. **HTTP with SSE transport**
559 | * Uses Server-Sent Events for server-to-client messages
560 | * HTTP POST for client-to-server messages
561 |
562 | All transports use [JSON-RPC](https://www.jsonrpc.org/) 2.0 to exchange messages. See the [specification](https://spec.modelcontextprotocol.io) for detailed information about the Model Context Protocol message format.
563 |
564 | ### Message types
565 |
566 | MCP has these main types of messages:
567 |
568 | 1. **Requests** expect a response from the other side:
569 | ```typescript
570 | interface Request {
571 | method: string;
572 | params?: { ... };
573 | }
574 | ```
575 |
576 | 2. **Results** are successful responses to requests:
577 | ```typescript
578 | interface Result {
579 | [key: string]: unknown;
580 | }
581 | ```
582 |
583 | 3. **Errors** indicate that a request failed:
584 | ```typescript
585 | interface Error {
586 | code: number;
587 | message: string;
588 | data?: unknown;
589 | }
590 | ```
591 |
592 | 4. **Notifications** are one-way messages that don't expect a response:
593 | ```typescript
594 | interface Notification {
595 | method: string;
596 | params?: { ... };
597 | }
598 | ```
599 |
600 | ## Connection lifecycle
601 |
602 | ### 1. Initialization
603 |
604 | ```mermaid
605 | sequenceDiagram
606 | participant Client
607 | participant Server
608 |
609 | Client->>Server: initialize request
610 | Server->>Client: initialize response
611 | Client->>Server: initialized notification
612 |
613 | Note over Client,Server: Connection ready for use
614 | ```
615 |
616 | 1. Client sends `initialize` request with protocol version and capabilities
617 | 2. Server responds with its protocol version and capabilities
618 | 3. Client sends `initialized` notification as acknowledgment
619 | 4. Normal message exchange begins
620 |
621 | ### 2. Message exchange
622 |
623 | After initialization, the following patterns are supported:
624 |
625 | * **Request-Response**: Client or server sends requests, the other responds
626 | * **Notifications**: Either party sends one-way messages
627 |
628 | ### 3. Termination
629 |
630 | Either party can terminate the connection:
631 |
632 | * Clean shutdown via `close()`
633 | * Transport disconnection
634 | * Error conditions
635 |
636 | ## Error handling
637 |
638 | MCP defines these standard error codes:
639 |
640 | ```typescript
641 | enum ErrorCode {
642 | // Standard JSON-RPC error codes
643 | ParseError = -32700,
644 | InvalidRequest = -32600,
645 | MethodNotFound = -32601,
646 | InvalidParams = -32602,
647 | InternalError = -32603
648 | }
649 | ```
650 |
651 | SDKs and applications can define their own error codes above -32000.
652 |
653 | Errors are propagated through:
654 |
655 | * Error responses to requests
656 | * Error events on transports
657 | * Protocol-level error handlers
658 |
659 | ## Implementation example
660 |
661 | Here's a basic example of implementing an MCP server:
662 |
663 | <Tabs>
664 | <Tab title="TypeScript">
665 | ```typescript
666 | import { Server } from "@modelcontextprotocol/sdk/server/index.js";
667 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
668 |
669 | const server = new Server({
670 | name: "example-server",
671 | version: "1.0.0"
672 | }, {
673 | capabilities: {
674 | resources: {}
675 | }
676 | });
677 |
678 | // Handle requests
679 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
680 | return {
681 | resources: [
682 | {
683 | uri: "example://resource",
684 | name: "Example Resource"
685 | }
686 | ]
687 | };
688 | });
689 |
690 | // Connect transport
691 | const transport = new StdioServerTransport();
692 | await server.connect(transport);
693 | ```
694 | </Tab>
695 |
696 | <Tab title="Python">
697 | ```python
698 | import asyncio
699 | import mcp.types as types
700 | from mcp.server import Server
701 | from mcp.server.stdio import stdio_server
702 |
703 | app = Server("example-server")
704 |
705 | @app.list_resources()
706 | async def list_resources() -> list[types.Resource]:
707 | return [
708 | types.Resource(
709 | uri="example://resource",
710 | name="Example Resource"
711 | )
712 | ]
713 |
714 | async def main():
715 | async with stdio_server() as streams:
716 | await app.run(
717 | streams[0],
718 | streams[1],
719 | app.create_initialization_options()
720 | )
721 |
722 | if __name__ == "__main__":
723 | asyncio.run(main)
724 | ```
725 | </Tab>
726 | </Tabs>
727 |
728 | ## Best practices
729 |
730 | ### Transport selection
731 |
732 | 1. **Local communication**
733 | * Use stdio transport for local processes
734 | * Efficient for same-machine communication
735 | * Simple process management
736 |
737 | 2. **Remote communication**
738 | * Use SSE for scenarios requiring HTTP compatibility
739 | * Consider security implications including authentication and authorization
740 |
741 | ### Message handling
742 |
743 | 1. **Request processing**
744 | * Validate inputs thoroughly
745 | * Use type-safe schemas
746 | * Handle errors gracefully
747 | * Implement timeouts
748 |
749 | 2. **Progress reporting**
750 | * Use progress tokens for long operations
751 | * Report progress incrementally
752 | * Include total progress when known
753 |
754 | 3. **Error management**
755 | * Use appropriate error codes
756 | * Include helpful error messages
757 | * Clean up resources on errors
758 |
759 | ## Security considerations
760 |
761 | 1. **Transport security**
762 | * Use TLS for remote connections
763 | * Validate connection origins
764 | * Implement authentication when needed
765 |
766 | 2. **Message validation**
767 | * Validate all incoming messages
768 | * Sanitize inputs
769 | * Check message size limits
770 | * Verify JSON-RPC format
771 |
772 | 3. **Resource protection**
773 | * Implement access controls
774 | * Validate resource paths
775 | * Monitor resource usage
776 | * Rate limit requests
777 |
778 | 4. **Error handling**
779 | * Don't leak sensitive information
780 | * Log security-relevant errors
781 | * Implement proper cleanup
782 | * Handle DoS scenarios
783 |
784 | ## Debugging and monitoring
785 |
786 | 1. **Logging**
787 | * Log protocol events
788 | * Track message flow
789 | * Monitor performance
790 | * Record errors
791 |
792 | 2. **Diagnostics**
793 | * Implement health checks
794 | * Monitor connection state
795 | * Track resource usage
796 | * Profile performance
797 |
798 | 3. **Testing**
799 | * Test different transports
800 | * Verify error handling
801 | * Check edge cases
802 | * Load test servers
803 |
804 |
805 | # Prompts
806 | Source: https://modelcontextprotocol.io/docs/concepts/prompts
807 |
808 | Create reusable prompt templates and workflows
809 |
810 | Prompts enable servers to define reusable prompt templates and workflows that clients can easily surface to users and LLMs. They provide a powerful way to standardize and share common LLM interactions.
811 |
812 | <Note>
813 | Prompts are designed to be **user-controlled**, meaning they are exposed from servers to clients with the intention of the user being able to explicitly select them for use.
814 | </Note>
815 |
816 | ## Overview
817 |
818 | Prompts in MCP are predefined templates that can:
819 |
820 | * Accept dynamic arguments
821 | * Include context from resources
822 | * Chain multiple interactions
823 | * Guide specific workflows
824 | * Surface as UI elements (like slash commands)
825 |
826 | ## Prompt structure
827 |
828 | Each prompt is defined with:
829 |
830 | ```typescript
831 | {
832 | name: string; // Unique identifier for the prompt
833 | description?: string; // Human-readable description
834 | arguments?: [ // Optional list of arguments
835 | {
836 | name: string; // Argument identifier
837 | description?: string; // Argument description
838 | required?: boolean; // Whether argument is required
839 | }
840 | ]
841 | }
842 | ```
843 |
844 | ## Discovering prompts
845 |
846 | Clients can discover available prompts through the `prompts/list` endpoint:
847 |
848 | ```typescript
849 | // Request
850 | {
851 | method: "prompts/list"
852 | }
853 |
854 | // Response
855 | {
856 | prompts: [
857 | {
858 | name: "analyze-code",
859 | description: "Analyze code for potential improvements",
860 | arguments: [
861 | {
862 | name: "language",
863 | description: "Programming language",
864 | required: true
865 | }
866 | ]
867 | }
868 | ]
869 | }
870 | ```
871 |
872 | ## Using prompts
873 |
874 | To use a prompt, clients make a `prompts/get` request:
875 |
876 | ````typescript
877 | // Request
878 | {
879 | method: "prompts/get",
880 | params: {
881 | name: "analyze-code",
882 | arguments: {
883 | language: "python"
884 | }
885 | }
886 | }
887 |
888 | // Response
889 | {
890 | description: "Analyze Python code for potential improvements",
891 | messages: [
892 | {
893 | role: "user",
894 | content: {
895 | type: "text",
896 | text: "Please analyze the following Python code for potential improvements:\n\n```python\ndef calculate_sum(numbers):\n total = 0\n for num in numbers:\n total = total + num\n return total\n\nresult = calculate_sum([1, 2, 3, 4, 5])\nprint(result)\n```"
897 | }
898 | }
899 | ]
900 | }
901 | ````
902 |
903 | ## Dynamic prompts
904 |
905 | Prompts can be dynamic and include:
906 |
907 | ### Embedded resource context
908 |
909 | ```json
910 | {
911 | "name": "analyze-project",
912 | "description": "Analyze project logs and code",
913 | "arguments": [
914 | {
915 | "name": "timeframe",
916 | "description": "Time period to analyze logs",
917 | "required": true
918 | },
919 | {
920 | "name": "fileUri",
921 | "description": "URI of code file to review",
922 | "required": true
923 | }
924 | ]
925 | }
926 | ```
927 |
928 | When handling the `prompts/get` request:
929 |
930 | ```json
931 | {
932 | "messages": [
933 | {
934 | "role": "user",
935 | "content": {
936 | "type": "text",
937 | "text": "Analyze these system logs and the code file for any issues:"
938 | }
939 | },
940 | {
941 | "role": "user",
942 | "content": {
943 | "type": "resource",
944 | "resource": {
945 | "uri": "logs://recent?timeframe=1h",
946 | "text": "[2024-03-14 15:32:11] ERROR: Connection timeout in network.py:127\n[2024-03-14 15:32:15] WARN: Retrying connection (attempt 2/3)\n[2024-03-14 15:32:20] ERROR: Max retries exceeded",
947 | "mimeType": "text/plain"
948 | }
949 | }
950 | },
951 | {
952 | "role": "user",
953 | "content": {
954 | "type": "resource",
955 | "resource": {
956 | "uri": "file:///path/to/code.py",
957 | "text": "def connect_to_service(timeout=30):\n retries = 3\n for attempt in range(retries):\n try:\n return establish_connection(timeout)\n except TimeoutError:\n if attempt == retries - 1:\n raise\n time.sleep(5)\n\ndef establish_connection(timeout):\n # Connection implementation\n pass",
958 | "mimeType": "text/x-python"
959 | }
960 | }
961 | }
962 | ]
963 | }
964 | ```
965 |
966 | ### Multi-step workflows
967 |
968 | ```typescript
969 | const debugWorkflow = {
970 | name: "debug-error",
971 | async getMessages(error: string) {
972 | return [
973 | {
974 | role: "user",
975 | content: {
976 | type: "text",
977 | text: `Here's an error I'm seeing: ${error}`
978 | }
979 | },
980 | {
981 | role: "assistant",
982 | content: {
983 | type: "text",
984 | text: "I'll help analyze this error. What have you tried so far?"
985 | }
986 | },
987 | {
988 | role: "user",
989 | content: {
990 | type: "text",
991 | text: "I've tried restarting the service, but the error persists."
992 | }
993 | }
994 | ];
995 | }
996 | };
997 | ```
998 |
999 | ## Example implementation
1000 |
1001 | Here's a complete example of implementing prompts in an MCP server:
1002 |
1003 | <Tabs>
1004 | <Tab title="TypeScript">
1005 | ```typescript
1006 | import { Server } from "@modelcontextprotocol/sdk/server";
1007 | import {
1008 | ListPromptsRequestSchema,
1009 | GetPromptRequestSchema
1010 | } from "@modelcontextprotocol/sdk/types";
1011 |
1012 | const PROMPTS = {
1013 | "git-commit": {
1014 | name: "git-commit",
1015 | description: "Generate a Git commit message",
1016 | arguments: [
1017 | {
1018 | name: "changes",
1019 | description: "Git diff or description of changes",
1020 | required: true
1021 | }
1022 | ]
1023 | },
1024 | "explain-code": {
1025 | name: "explain-code",
1026 | description: "Explain how code works",
1027 | arguments: [
1028 | {
1029 | name: "code",
1030 | description: "Code to explain",
1031 | required: true
1032 | },
1033 | {
1034 | name: "language",
1035 | description: "Programming language",
1036 | required: false
1037 | }
1038 | ]
1039 | }
1040 | };
1041 |
1042 | const server = new Server({
1043 | name: "example-prompts-server",
1044 | version: "1.0.0"
1045 | }, {
1046 | capabilities: {
1047 | prompts: {}
1048 | }
1049 | });
1050 |
1051 | // List available prompts
1052 | server.setRequestHandler(ListPromptsRequestSchema, async () => {
1053 | return {
1054 | prompts: Object.values(PROMPTS)
1055 | };
1056 | });
1057 |
1058 | // Get specific prompt
1059 | server.setRequestHandler(GetPromptRequestSchema, async (request) => {
1060 | const prompt = PROMPTS[request.params.name];
1061 | if (!prompt) {
1062 | throw new Error(`Prompt not found: ${request.params.name}`);
1063 | }
1064 |
1065 | if (request.params.name === "git-commit") {
1066 | return {
1067 | messages: [
1068 | {
1069 | role: "user",
1070 | content: {
1071 | type: "text",
1072 | text: `Generate a concise but descriptive commit message for these changes:\n\n${request.params.arguments?.changes}`
1073 | }
1074 | }
1075 | ]
1076 | };
1077 | }
1078 |
1079 | if (request.params.name === "explain-code") {
1080 | const language = request.params.arguments?.language || "Unknown";
1081 | return {
1082 | messages: [
1083 | {
1084 | role: "user",
1085 | content: {
1086 | type: "text",
1087 | text: `Explain how this ${language} code works:\n\n${request.params.arguments?.code}`
1088 | }
1089 | }
1090 | ]
1091 | };
1092 | }
1093 |
1094 | throw new Error("Prompt implementation not found");
1095 | });
1096 | ```
1097 | </Tab>
1098 |
1099 | <Tab title="Python">
1100 | ```python
1101 | from mcp.server import Server
1102 | import mcp.types as types
1103 |
1104 | # Define available prompts
1105 | PROMPTS = {
1106 | "git-commit": types.Prompt(
1107 | name="git-commit",
1108 | description="Generate a Git commit message",
1109 | arguments=[
1110 | types.PromptArgument(
1111 | name="changes",
1112 | description="Git diff or description of changes",
1113 | required=True
1114 | )
1115 | ],
1116 | ),
1117 | "explain-code": types.Prompt(
1118 | name="explain-code",
1119 | description="Explain how code works",
1120 | arguments=[
1121 | types.PromptArgument(
1122 | name="code",
1123 | description="Code to explain",
1124 | required=True
1125 | ),
1126 | types.PromptArgument(
1127 | name="language",
1128 | description="Programming language",
1129 | required=False
1130 | )
1131 | ],
1132 | )
1133 | }
1134 |
1135 | # Initialize server
1136 | app = Server("example-prompts-server")
1137 |
1138 | @app.list_prompts()
1139 | async def list_prompts() -> list[types.Prompt]:
1140 | return list(PROMPTS.values())
1141 |
1142 | @app.get_prompt()
1143 | async def get_prompt(
1144 | name: str, arguments: dict[str, str] | None = None
1145 | ) -> types.GetPromptResult:
1146 | if name not in PROMPTS:
1147 | raise ValueError(f"Prompt not found: {name}")
1148 |
1149 | if name == "git-commit":
1150 | changes = arguments.get("changes") if arguments else ""
1151 | return types.GetPromptResult(
1152 | messages=[
1153 | types.PromptMessage(
1154 | role="user",
1155 | content=types.TextContent(
1156 | type="text",
1157 | text=f"Generate a concise but descriptive commit message "
1158 | f"for these changes:\n\n{changes}"
1159 | )
1160 | )
1161 | ]
1162 | )
1163 |
1164 | if name == "explain-code":
1165 | code = arguments.get("code") if arguments else ""
1166 | language = arguments.get("language", "Unknown") if arguments else "Unknown"
1167 | return types.GetPromptResult(
1168 | messages=[
1169 | types.PromptMessage(
1170 | role="user",
1171 | content=types.TextContent(
1172 | type="text",
1173 | text=f"Explain how this {language} code works:\n\n{code}"
1174 | )
1175 | )
1176 | ]
1177 | )
1178 |
1179 | raise ValueError("Prompt implementation not found")
1180 | ```
1181 | </Tab>
1182 | </Tabs>
1183 |
1184 | ## Best practices
1185 |
1186 | When implementing prompts:
1187 |
1188 | 1. Use clear, descriptive prompt names
1189 | 2. Provide detailed descriptions for prompts and arguments
1190 | 3. Validate all required arguments
1191 | 4. Handle missing arguments gracefully
1192 | 5. Consider versioning for prompt templates
1193 | 6. Cache dynamic content when appropriate
1194 | 7. Implement error handling
1195 | 8. Document expected argument formats
1196 | 9. Consider prompt composability
1197 | 10. Test prompts with various inputs
1198 |
1199 | ## UI integration
1200 |
1201 | Prompts can be surfaced in client UIs as:
1202 |
1203 | * Slash commands
1204 | * Quick actions
1205 | * Context menu items
1206 | * Command palette entries
1207 | * Guided workflows
1208 | * Interactive forms
1209 |
1210 | ## Updates and changes
1211 |
1212 | Servers can notify clients about prompt changes:
1213 |
1214 | 1. Server capability: `prompts.listChanged`
1215 | 2. Notification: `notifications/prompts/list_changed`
1216 | 3. Client re-fetches prompt list
1217 |
1218 | ## Security considerations
1219 |
1220 | When implementing prompts:
1221 |
1222 | * Validate all arguments
1223 | * Sanitize user input
1224 | * Consider rate limiting
1225 | * Implement access controls
1226 | * Audit prompt usage
1227 | * Handle sensitive data appropriately
1228 | * Validate generated content
1229 | * Implement timeouts
1230 | * Consider prompt injection risks
1231 | * Document security requirements
1232 |
1233 |
1234 | # Resources
1235 | Source: https://modelcontextprotocol.io/docs/concepts/resources
1236 |
1237 | Expose data and content from your servers to LLMs
1238 |
1239 | Resources are a core primitive in the Model Context Protocol (MCP) that allow servers to expose data and content that can be read by clients and used as context for LLM interactions.
1240 |
1241 | <Note>
1242 | Resources are designed to be **application-controlled**, meaning that the client application can decide how and when they should be used.
1243 | Different MCP clients may handle resources differently. For example:
1244 |
1245 | * Claude Desktop currently requires users to explicitly select resources before they can be used
1246 | * Other clients might automatically select resources based on heuristics
1247 | * Some implementations may even allow the AI model itself to determine which resources to use
1248 |
1249 | Server authors should be prepared to handle any of these interaction patterns when implementing resource support. In order to expose data to models automatically, server authors should use a **model-controlled** primitive such as [Tools](./tools).
1250 | </Note>
1251 |
1252 | ## Overview
1253 |
1254 | Resources represent any kind of data that an MCP server wants to make available to clients. This can include:
1255 |
1256 | * File contents
1257 | * Database records
1258 | * API responses
1259 | * Live system data
1260 | * Screenshots and images
1261 | * Log files
1262 | * And more
1263 |
1264 | Each resource is identified by a unique URI and can contain either text or binary data.
1265 |
1266 | ## Resource URIs
1267 |
1268 | Resources are identified using URIs that follow this format:
1269 |
1270 | ```
1271 | [protocol]://[host]/[path]
1272 | ```
1273 |
1274 | For example:
1275 |
1276 | * `file:///home/user/documents/report.pdf`
1277 | * `postgres://database/customers/schema`
1278 | * `screen://localhost/display1`
1279 |
1280 | The protocol and path structure is defined by the MCP server implementation. Servers can define their own custom URI schemes.
1281 |
1282 | ## Resource types
1283 |
1284 | Resources can contain two types of content:
1285 |
1286 | ### Text resources
1287 |
1288 | Text resources contain UTF-8 encoded text data. These are suitable for:
1289 |
1290 | * Source code
1291 | * Configuration files
1292 | * Log files
1293 | * JSON/XML data
1294 | * Plain text
1295 |
1296 | ### Binary resources
1297 |
1298 | Binary resources contain raw binary data encoded in base64. These are suitable for:
1299 |
1300 | * Images
1301 | * PDFs
1302 | * Audio files
1303 | * Video files
1304 | * Other non-text formats
1305 |
1306 | ## Resource discovery
1307 |
1308 | Clients can discover available resources through two main methods:
1309 |
1310 | ### Direct resources
1311 |
1312 | Servers expose a list of concrete resources via the `resources/list` endpoint. Each resource includes:
1313 |
1314 | ```typescript
1315 | {
1316 | uri: string; // Unique identifier for the resource
1317 | name: string; // Human-readable name
1318 | description?: string; // Optional description
1319 | mimeType?: string; // Optional MIME type
1320 | }
1321 | ```
1322 |
1323 | ### Resource templates
1324 |
1325 | For dynamic resources, servers can expose [URI templates](https://datatracker.ietf.org/doc/html/rfc6570) that clients can use to construct valid resource URIs:
1326 |
1327 | ```typescript
1328 | {
1329 | uriTemplate: string; // URI template following RFC 6570
1330 | name: string; // Human-readable name for this type
1331 | description?: string; // Optional description
1332 | mimeType?: string; // Optional MIME type for all matching resources
1333 | }
1334 | ```
1335 |
1336 | ## Reading resources
1337 |
1338 | To read a resource, clients make a `resources/read` request with the resource URI.
1339 |
1340 | The server responds with a list of resource contents:
1341 |
1342 | ```typescript
1343 | {
1344 | contents: [
1345 | {
1346 | uri: string; // The URI of the resource
1347 | mimeType?: string; // Optional MIME type
1348 |
1349 | // One of:
1350 | text?: string; // For text resources
1351 | blob?: string; // For binary resources (base64 encoded)
1352 | }
1353 | ]
1354 | }
1355 | ```
1356 |
1357 | <Tip>
1358 | Servers may return multiple resources in response to one `resources/read` request. This could be used, for example, to return a list of files inside a directory when the directory is read.
1359 | </Tip>
1360 |
1361 | ## Resource updates
1362 |
1363 | MCP supports real-time updates for resources through two mechanisms:
1364 |
1365 | ### List changes
1366 |
1367 | Servers can notify clients when their list of available resources changes via the `notifications/resources/list_changed` notification.
1368 |
1369 | ### Content changes
1370 |
1371 | Clients can subscribe to updates for specific resources:
1372 |
1373 | 1. Client sends `resources/subscribe` with resource URI
1374 | 2. Server sends `notifications/resources/updated` when the resource changes
1375 | 3. Client can fetch latest content with `resources/read`
1376 | 4. Client can unsubscribe with `resources/unsubscribe`
1377 |
1378 | ## Example implementation
1379 |
1380 | Here's a simple example of implementing resource support in an MCP server:
1381 |
1382 | <Tabs>
1383 | <Tab title="TypeScript">
1384 | ```typescript
1385 | const server = new Server({
1386 | name: "example-server",
1387 | version: "1.0.0"
1388 | }, {
1389 | capabilities: {
1390 | resources: {}
1391 | }
1392 | });
1393 |
1394 | // List available resources
1395 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
1396 | return {
1397 | resources: [
1398 | {
1399 | uri: "file:///logs/app.log",
1400 | name: "Application Logs",
1401 | mimeType: "text/plain"
1402 | }
1403 | ]
1404 | };
1405 | });
1406 |
1407 | // Read resource contents
1408 | server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
1409 | const uri = request.params.uri;
1410 |
1411 | if (uri === "file:///logs/app.log") {
1412 | const logContents = await readLogFile();
1413 | return {
1414 | contents: [
1415 | {
1416 | uri,
1417 | mimeType: "text/plain",
1418 | text: logContents
1419 | }
1420 | ]
1421 | };
1422 | }
1423 |
1424 | throw new Error("Resource not found");
1425 | });
1426 | ```
1427 | </Tab>
1428 |
1429 | <Tab title="Python">
1430 | ```python
1431 | app = Server("example-server")
1432 |
1433 | @app.list_resources()
1434 | async def list_resources() -> list[types.Resource]:
1435 | return [
1436 | types.Resource(
1437 | uri="file:///logs/app.log",
1438 | name="Application Logs",
1439 | mimeType="text/plain"
1440 | )
1441 | ]
1442 |
1443 | @app.read_resource()
1444 | async def read_resource(uri: AnyUrl) -> str:
1445 | if str(uri) == "file:///logs/app.log":
1446 | log_contents = await read_log_file()
1447 | return log_contents
1448 |
1449 | raise ValueError("Resource not found")
1450 |
1451 | # Start server
1452 | async with stdio_server() as streams:
1453 | await app.run(
1454 | streams[0],
1455 | streams[1],
1456 | app.create_initialization_options()
1457 | )
1458 | ```
1459 | </Tab>
1460 | </Tabs>
1461 |
1462 | ## Best practices
1463 |
1464 | When implementing resource support:
1465 |
1466 | 1. Use clear, descriptive resource names and URIs
1467 | 2. Include helpful descriptions to guide LLM understanding
1468 | 3. Set appropriate MIME types when known
1469 | 4. Implement resource templates for dynamic content
1470 | 5. Use subscriptions for frequently changing resources
1471 | 6. Handle errors gracefully with clear error messages
1472 | 7. Consider pagination for large resource lists
1473 | 8. Cache resource contents when appropriate
1474 | 9. Validate URIs before processing
1475 | 10. Document your custom URI schemes
1476 |
1477 | ## Security considerations
1478 |
1479 | When exposing resources:
1480 |
1481 | * Validate all resource URIs
1482 | * Implement appropriate access controls
1483 | * Sanitize file paths to prevent directory traversal
1484 | * Be cautious with binary data handling
1485 | * Consider rate limiting for resource reads
1486 | * Audit resource access
1487 | * Encrypt sensitive data in transit
1488 | * Validate MIME types
1489 | * Implement timeouts for long-running reads
1490 | * Handle resource cleanup appropriately
1491 |
1492 |
1493 | # Roots
1494 | Source: https://modelcontextprotocol.io/docs/concepts/roots
1495 |
1496 | Understanding roots in MCP
1497 |
1498 | Roots are a concept in MCP that define the boundaries where servers can operate. They provide a way for clients to inform servers about relevant resources and their locations.
1499 |
1500 | ## What are Roots?
1501 |
1502 | A root is a URI that a client suggests a server should focus on. When a client connects to a server, it declares which roots the server should work with. While primarily used for filesystem paths, roots can be any valid URI including HTTP URLs.
1503 |
1504 | For example, roots could be:
1505 |
1506 | ```
1507 | file:///home/user/projects/myapp
1508 | https://api.example.com/v1
1509 | ```
1510 |
1511 | ## Why Use Roots?
1512 |
1513 | Roots serve several important purposes:
1514 |
1515 | 1. **Guidance**: They inform servers about relevant resources and locations
1516 | 2. **Clarity**: Roots make it clear which resources are part of your workspace
1517 | 3. **Organization**: Multiple roots let you work with different resources simultaneously
1518 |
1519 | ## How Roots Work
1520 |
1521 | When a client supports roots, it:
1522 |
1523 | 1. Declares the `roots` capability during connection
1524 | 2. Provides a list of suggested roots to the server
1525 | 3. Notifies the server when roots change (if supported)
1526 |
1527 | While roots are informational and not strictly enforcing, servers should:
1528 |
1529 | 1. Respect the provided roots
1530 | 2. Use root URIs to locate and access resources
1531 | 3. Prioritize operations within root boundaries
1532 |
1533 | ## Common Use Cases
1534 |
1535 | Roots are commonly used to define:
1536 |
1537 | * Project directories
1538 | * Repository locations
1539 | * API endpoints
1540 | * Configuration locations
1541 | * Resource boundaries
1542 |
1543 | ## Best Practices
1544 |
1545 | When working with roots:
1546 |
1547 | 1. Only suggest necessary resources
1548 | 2. Use clear, descriptive names for roots
1549 | 3. Monitor root accessibility
1550 | 4. Handle root changes gracefully
1551 |
1552 | ## Example
1553 |
1554 | Here's how a typical MCP client might expose roots:
1555 |
1556 | ```json
1557 | {
1558 | "roots": [
1559 | {
1560 | "uri": "file:///home/user/projects/frontend",
1561 | "name": "Frontend Repository"
1562 | },
1563 | {
1564 | "uri": "https://api.example.com/v1",
1565 | "name": "API Endpoint"
1566 | }
1567 | ]
1568 | }
1569 | ```
1570 |
1571 | This configuration suggests the server focus on both a local repository and an API endpoint while keeping them logically separated.
1572 |
1573 |
1574 | # Sampling
1575 | Source: https://modelcontextprotocol.io/docs/concepts/sampling
1576 |
1577 | Let your servers request completions from LLMs
1578 |
1579 | Sampling is a powerful MCP feature that allows servers to request LLM completions through the client, enabling sophisticated agentic behaviors while maintaining security and privacy.
1580 |
1581 | <Info>
1582 | This feature of MCP is not yet supported in the Claude Desktop client.
1583 | </Info>
1584 |
1585 | ## How sampling works
1586 |
1587 | The sampling flow follows these steps:
1588 |
1589 | 1. Server sends a `sampling/createMessage` request to the client
1590 | 2. Client reviews the request and can modify it
1591 | 3. Client samples from an LLM
1592 | 4. Client reviews the completion
1593 | 5. Client returns the result to the server
1594 |
1595 | This human-in-the-loop design ensures users maintain control over what the LLM sees and generates.
1596 |
1597 | ## Message format
1598 |
1599 | Sampling requests use a standardized message format:
1600 |
1601 | ```typescript
1602 | {
1603 | messages: [
1604 | {
1605 | role: "user" | "assistant",
1606 | content: {
1607 | type: "text" | "image",
1608 |
1609 | // For text:
1610 | text?: string,
1611 |
1612 | // For images:
1613 | data?: string, // base64 encoded
1614 | mimeType?: string
1615 | }
1616 | }
1617 | ],
1618 | modelPreferences?: {
1619 | hints?: [{
1620 | name?: string // Suggested model name/family
1621 | }],
1622 | costPriority?: number, // 0-1, importance of minimizing cost
1623 | speedPriority?: number, // 0-1, importance of low latency
1624 | intelligencePriority?: number // 0-1, importance of capabilities
1625 | },
1626 | systemPrompt?: string,
1627 | includeContext?: "none" | "thisServer" | "allServers",
1628 | temperature?: number,
1629 | maxTokens: number,
1630 | stopSequences?: string[],
1631 | metadata?: Record<string, unknown>
1632 | }
1633 | ```
1634 |
1635 | ## Request parameters
1636 |
1637 | ### Messages
1638 |
1639 | The `messages` array contains the conversation history to send to the LLM. Each message has:
1640 |
1641 | * `role`: Either "user" or "assistant"
1642 | * `content`: The message content, which can be:
1643 | * Text content with a `text` field
1644 | * Image content with `data` (base64) and `mimeType` fields
1645 |
1646 | ### Model preferences
1647 |
1648 | The `modelPreferences` object allows servers to specify their model selection preferences:
1649 |
1650 | * `hints`: Array of model name suggestions that clients can use to select an appropriate model:
1651 | * `name`: String that can match full or partial model names (e.g. "claude-3", "sonnet")
1652 | * Clients may map hints to equivalent models from different providers
1653 | * Multiple hints are evaluated in preference order
1654 |
1655 | * Priority values (0-1 normalized):
1656 | * `costPriority`: Importance of minimizing costs
1657 | * `speedPriority`: Importance of low latency response
1658 | * `intelligencePriority`: Importance of advanced model capabilities
1659 |
1660 | Clients make the final model selection based on these preferences and their available models.
1661 |
1662 | ### System prompt
1663 |
1664 | An optional `systemPrompt` field allows servers to request a specific system prompt. The client may modify or ignore this.
1665 |
1666 | ### Context inclusion
1667 |
1668 | The `includeContext` parameter specifies what MCP context to include:
1669 |
1670 | * `"none"`: No additional context
1671 | * `"thisServer"`: Include context from the requesting server
1672 | * `"allServers"`: Include context from all connected MCP servers
1673 |
1674 | The client controls what context is actually included.
1675 |
1676 | ### Sampling parameters
1677 |
1678 | Fine-tune the LLM sampling with:
1679 |
1680 | * `temperature`: Controls randomness (0.0 to 1.0)
1681 | * `maxTokens`: Maximum tokens to generate
1682 | * `stopSequences`: Array of sequences that stop generation
1683 | * `metadata`: Additional provider-specific parameters
1684 |
1685 | ## Response format
1686 |
1687 | The client returns a completion result:
1688 |
1689 | ```typescript
1690 | {
1691 | model: string, // Name of the model used
1692 | stopReason?: "endTurn" | "stopSequence" | "maxTokens" | string,
1693 | role: "user" | "assistant",
1694 | content: {
1695 | type: "text" | "image",
1696 | text?: string,
1697 | data?: string,
1698 | mimeType?: string
1699 | }
1700 | }
1701 | ```
1702 |
1703 | ## Example request
1704 |
1705 | Here's an example of requesting sampling from a client:
1706 |
1707 | ```json
1708 | {
1709 | "method": "sampling/createMessage",
1710 | "params": {
1711 | "messages": [
1712 | {
1713 | "role": "user",
1714 | "content": {
1715 | "type": "text",
1716 | "text": "What files are in the current directory?"
1717 | }
1718 | }
1719 | ],
1720 | "systemPrompt": "You are a helpful file system assistant.",
1721 | "includeContext": "thisServer",
1722 | "maxTokens": 100
1723 | }
1724 | }
1725 | ```
1726 |
1727 | ## Best practices
1728 |
1729 | When implementing sampling:
1730 |
1731 | 1. Always provide clear, well-structured prompts
1732 | 2. Handle both text and image content appropriately
1733 | 3. Set reasonable token limits
1734 | 4. Include relevant context through `includeContext`
1735 | 5. Validate responses before using them
1736 | 6. Handle errors gracefully
1737 | 7. Consider rate limiting sampling requests
1738 | 8. Document expected sampling behavior
1739 | 9. Test with various model parameters
1740 | 10. Monitor sampling costs
1741 |
1742 | ## Human in the loop controls
1743 |
1744 | Sampling is designed with human oversight in mind:
1745 |
1746 | ### For prompts
1747 |
1748 | * Clients should show users the proposed prompt
1749 | * Users should be able to modify or reject prompts
1750 | * System prompts can be filtered or modified
1751 | * Context inclusion is controlled by the client
1752 |
1753 | ### For completions
1754 |
1755 | * Clients should show users the completion
1756 | * Users should be able to modify or reject completions
1757 | * Clients can filter or modify completions
1758 | * Users control which model is used
1759 |
1760 | ## Security considerations
1761 |
1762 | When implementing sampling:
1763 |
1764 | * Validate all message content
1765 | * Sanitize sensitive information
1766 | * Implement appropriate rate limits
1767 | * Monitor sampling usage
1768 | * Encrypt data in transit
1769 | * Handle user data privacy
1770 | * Audit sampling requests
1771 | * Control cost exposure
1772 | * Implement timeouts
1773 | * Handle model errors gracefully
1774 |
1775 | ## Common patterns
1776 |
1777 | ### Agentic workflows
1778 |
1779 | Sampling enables agentic patterns like:
1780 |
1781 | * Reading and analyzing resources
1782 | * Making decisions based on context
1783 | * Generating structured data
1784 | * Handling multi-step tasks
1785 | * Providing interactive assistance
1786 |
1787 | ### Context management
1788 |
1789 | Best practices for context:
1790 |
1791 | * Request minimal necessary context
1792 | * Structure context clearly
1793 | * Handle context size limits
1794 | * Update context as needed
1795 | * Clean up stale context
1796 |
1797 | ### Error handling
1798 |
1799 | Robust error handling should:
1800 |
1801 | * Catch sampling failures
1802 | * Handle timeout errors
1803 | * Manage rate limits
1804 | * Validate responses
1805 | * Provide fallback behaviors
1806 | * Log errors appropriately
1807 |
1808 | ## Limitations
1809 |
1810 | Be aware of these limitations:
1811 |
1812 | * Sampling depends on client capabilities
1813 | * Users control sampling behavior
1814 | * Context size has limits
1815 | * Rate limits may apply
1816 | * Costs should be considered
1817 | * Model availability varies
1818 | * Response times vary
1819 | * Not all content types supported
1820 |
1821 |
1822 | # Tools
1823 | Source: https://modelcontextprotocol.io/docs/concepts/tools
1824 |
1825 | Enable LLMs to perform actions through your server
1826 |
1827 | Tools are a powerful primitive in the Model Context Protocol (MCP) that enable servers to expose executable functionality to clients. Through tools, LLMs can interact with external systems, perform computations, and take actions in the real world.
1828 |
1829 | <Note>
1830 | Tools are designed to be **model-controlled**, meaning that tools are exposed from servers to clients with the intention of the AI model being able to automatically invoke them (with a human in the loop to grant approval).
1831 | </Note>
1832 |
1833 | ## Overview
1834 |
1835 | Tools in MCP allow servers to expose executable functions that can be invoked by clients and used by LLMs to perform actions. Key aspects of tools include:
1836 |
1837 | * **Discovery**: Clients can list available tools through the `tools/list` endpoint
1838 | * **Invocation**: Tools are called using the `tools/call` endpoint, where servers perform the requested operation and return results
1839 | * **Flexibility**: Tools can range from simple calculations to complex API interactions
1840 |
1841 | Like [resources](/docs/concepts/resources), tools are identified by unique names and can include descriptions to guide their usage. However, unlike resources, tools represent dynamic operations that can modify state or interact with external systems.
1842 |
1843 | ## Tool definition structure
1844 |
1845 | Each tool is defined with the following structure:
1846 |
1847 | ```typescript
1848 | {
1849 | name: string; // Unique identifier for the tool
1850 | description?: string; // Human-readable description
1851 | inputSchema: { // JSON Schema for the tool's parameters
1852 | type: "object",
1853 | properties: { ... } // Tool-specific parameters
1854 | }
1855 | }
1856 | ```
1857 |
1858 | ## Implementing tools
1859 |
1860 | Here's an example of implementing a basic tool in an MCP server:
1861 |
1862 | <Tabs>
1863 | <Tab title="TypeScript">
1864 | ```typescript
1865 | const server = new Server({
1866 | name: "example-server",
1867 | version: "1.0.0"
1868 | }, {
1869 | capabilities: {
1870 | tools: {}
1871 | }
1872 | });
1873 |
1874 | // Define available tools
1875 | server.setRequestHandler(ListToolsRequestSchema, async () => {
1876 | return {
1877 | tools: [{
1878 | name: "calculate_sum",
1879 | description: "Add two numbers together",
1880 | inputSchema: {
1881 | type: "object",
1882 | properties: {
1883 | a: { type: "number" },
1884 | b: { type: "number" }
1885 | },
1886 | required: ["a", "b"]
1887 | }
1888 | }]
1889 | };
1890 | });
1891 |
1892 | // Handle tool execution
1893 | server.setRequestHandler(CallToolRequestSchema, async (request) => {
1894 | if (request.params.name === "calculate_sum") {
1895 | const { a, b } = request.params.arguments;
1896 | return {
1897 | content: [
1898 | {
1899 | type: "text",
1900 | text: String(a + b)
1901 | }
1902 | ]
1903 | };
1904 | }
1905 | throw new Error("Tool not found");
1906 | });
1907 | ```
1908 | </Tab>
1909 |
1910 | <Tab title="Python">
1911 | ```python
1912 | app = Server("example-server")
1913 |
1914 | @app.list_tools()
1915 | async def list_tools() -> list[types.Tool]:
1916 | return [
1917 | types.Tool(
1918 | name="calculate_sum",
1919 | description="Add two numbers together",
1920 | inputSchema={
1921 | "type": "object",
1922 | "properties": {
1923 | "a": {"type": "number"},
1924 | "b": {"type": "number"}
1925 | },
1926 | "required": ["a", "b"]
1927 | }
1928 | )
1929 | ]
1930 |
1931 | @app.call_tool()
1932 | async def call_tool(
1933 | name: str,
1934 | arguments: dict
1935 | ) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
1936 | if name == "calculate_sum":
1937 | a = arguments["a"]
1938 | b = arguments["b"]
1939 | result = a + b
1940 | return [types.TextContent(type="text", text=str(result))]
1941 | raise ValueError(f"Tool not found: {name}")
1942 | ```
1943 | </Tab>
1944 | </Tabs>
1945 |
1946 | ## Example tool patterns
1947 |
1948 | Here are some examples of types of tools that a server could provide:
1949 |
1950 | ### System operations
1951 |
1952 | Tools that interact with the local system:
1953 |
1954 | ```typescript
1955 | {
1956 | name: "execute_command",
1957 | description: "Run a shell command",
1958 | inputSchema: {
1959 | type: "object",
1960 | properties: {
1961 | command: { type: "string" },
1962 | args: { type: "array", items: { type: "string" } }
1963 | }
1964 | }
1965 | }
1966 | ```
1967 |
1968 | ### API integrations
1969 |
1970 | Tools that wrap external APIs:
1971 |
1972 | ```typescript
1973 | {
1974 | name: "github_create_issue",
1975 | description: "Create a GitHub issue",
1976 | inputSchema: {
1977 | type: "object",
1978 | properties: {
1979 | title: { type: "string" },
1980 | body: { type: "string" },
1981 | labels: { type: "array", items: { type: "string" } }
1982 | }
1983 | }
1984 | }
1985 | ```
1986 |
1987 | ### Data processing
1988 |
1989 | Tools that transform or analyze data:
1990 |
1991 | ```typescript
1992 | {
1993 | name: "analyze_csv",
1994 | description: "Analyze a CSV file",
1995 | inputSchema: {
1996 | type: "object",
1997 | properties: {
1998 | filepath: { type: "string" },
1999 | operations: {
2000 | type: "array",
2001 | items: {
2002 | enum: ["sum", "average", "count"]
2003 | }
2004 | }
2005 | }
2006 | }
2007 | }
2008 | ```
2009 |
2010 | ## Best practices
2011 |
2012 | When implementing tools:
2013 |
2014 | 1. Provide clear, descriptive names and descriptions
2015 | 2. Use detailed JSON Schema definitions for parameters
2016 | 3. Include examples in tool descriptions to demonstrate how the model should use them
2017 | 4. Implement proper error handling and validation
2018 | 5. Use progress reporting for long operations
2019 | 6. Keep tool operations focused and atomic
2020 | 7. Document expected return value structures
2021 | 8. Implement proper timeouts
2022 | 9. Consider rate limiting for resource-intensive operations
2023 | 10. Log tool usage for debugging and monitoring
2024 |
2025 | ## Security considerations
2026 |
2027 | When exposing tools:
2028 |
2029 | ### Input validation
2030 |
2031 | * Validate all parameters against the schema
2032 | * Sanitize file paths and system commands
2033 | * Validate URLs and external identifiers
2034 | * Check parameter sizes and ranges
2035 | * Prevent command injection
2036 |
2037 | ### Access control
2038 |
2039 | * Implement authentication where needed
2040 | * Use appropriate authorization checks
2041 | * Audit tool usage
2042 | * Rate limit requests
2043 | * Monitor for abuse
2044 |
2045 | ### Error handling
2046 |
2047 | * Don't expose internal errors to clients
2048 | * Log security-relevant errors
2049 | * Handle timeouts appropriately
2050 | * Clean up resources after errors
2051 | * Validate return values
2052 |
2053 | ## Tool discovery and updates
2054 |
2055 | MCP supports dynamic tool discovery:
2056 |
2057 | 1. Clients can list available tools at any time
2058 | 2. Servers can notify clients when tools change using `notifications/tools/list_changed`
2059 | 3. Tools can be added or removed during runtime
2060 | 4. Tool definitions can be updated (though this should be done carefully)
2061 |
2062 | ## Error handling
2063 |
2064 | Tool errors should be reported within the result object, not as MCP protocol-level errors. This allows the LLM to see and potentially handle the error. When a tool encounters an error:
2065 |
2066 | 1. Set `isError` to `true` in the result
2067 | 2. Include error details in the `content` array
2068 |
2069 | Here's an example of proper error handling for tools:
2070 |
2071 | <Tabs>
2072 | <Tab title="TypeScript">
2073 | ```typescript
2074 | try {
2075 | // Tool operation
2076 | const result = performOperation();
2077 | return {
2078 | content: [
2079 | {
2080 | type: "text",
2081 | text: `Operation successful: ${result}`
2082 | }
2083 | ]
2084 | };
2085 | } catch (error) {
2086 | return {
2087 | isError: true,
2088 | content: [
2089 | {
2090 | type: "text",
2091 | text: `Error: ${error.message}`
2092 | }
2093 | ]
2094 | };
2095 | }
2096 | ```
2097 | </Tab>
2098 |
2099 | <Tab title="Python">
2100 | ```python
2101 | try:
2102 | # Tool operation
2103 | result = perform_operation()
2104 | return types.CallToolResult(
2105 | content=[
2106 | types.TextContent(
2107 | type="text",
2108 | text=f"Operation successful: {result}"
2109 | )
2110 | ]
2111 | )
2112 | except Exception as error:
2113 | return types.CallToolResult(
2114 | isError=True,
2115 | content=[
2116 | types.TextContent(
2117 | type="text",
2118 | text=f"Error: {str(error)}"
2119 | )
2120 | ]
2121 | )
2122 | ```
2123 | </Tab>
2124 | </Tabs>
2125 |
2126 | This approach allows the LLM to see that an error occurred and potentially take corrective action or request human intervention.
2127 |
2128 | ## Testing tools
2129 |
2130 | A comprehensive testing strategy for MCP tools should cover:
2131 |
2132 | * **Functional testing**: Verify tools execute correctly with valid inputs and handle invalid inputs appropriately
2133 | * **Integration testing**: Test tool interaction with external systems using both real and mocked dependencies
2134 | * **Security testing**: Validate authentication, authorization, input sanitization, and rate limiting
2135 | * **Performance testing**: Check behavior under load, timeout handling, and resource cleanup
2136 | * **Error handling**: Ensure tools properly report errors through the MCP protocol and clean up resources
2137 |
2138 |
2139 | # Transports
2140 | Source: https://modelcontextprotocol.io/docs/concepts/transports
2141 |
2142 | Learn about MCP's communication mechanisms
2143 |
2144 | Transports in the Model Context Protocol (MCP) provide the foundation for communication between clients and servers. A transport handles the underlying mechanics of how messages are sent and received.
2145 |
2146 | ## Message Format
2147 |
2148 | MCP uses [JSON-RPC](https://www.jsonrpc.org/) 2.0 as its wire format. The transport layer is responsible for converting MCP protocol messages into JSON-RPC format for transmission and converting received JSON-RPC messages back into MCP protocol messages.
2149 |
2150 | There are three types of JSON-RPC messages used:
2151 |
2152 | ### Requests
2153 |
2154 | ```typescript
2155 | {
2156 | jsonrpc: "2.0",
2157 | id: number | string,
2158 | method: string,
2159 | params?: object
2160 | }
2161 | ```
2162 |
2163 | ### Responses
2164 |
2165 | ```typescript
2166 | {
2167 | jsonrpc: "2.0",
2168 | id: number | string,
2169 | result?: object,
2170 | error?: {
2171 | code: number,
2172 | message: string,
2173 | data?: unknown
2174 | }
2175 | }
2176 | ```
2177 |
2178 | ### Notifications
2179 |
2180 | ```typescript
2181 | {
2182 | jsonrpc: "2.0",
2183 | method: string,
2184 | params?: object
2185 | }
2186 | ```
2187 |
2188 | ## Built-in Transport Types
2189 |
2190 | MCP includes two standard transport implementations:
2191 |
2192 | ### Standard Input/Output (stdio)
2193 |
2194 | The stdio transport enables communication through standard input and output streams. This is particularly useful for local integrations and command-line tools.
2195 |
2196 | Use stdio when:
2197 |
2198 | * Building command-line tools
2199 | * Implementing local integrations
2200 | * Needing simple process communication
2201 | * Working with shell scripts
2202 |
2203 | <Tabs>
2204 | <Tab title="TypeScript (Server)">
2205 | ```typescript
2206 | const server = new Server({
2207 | name: "example-server",
2208 | version: "1.0.0"
2209 | }, {
2210 | capabilities: {}
2211 | });
2212 |
2213 | const transport = new StdioServerTransport();
2214 | await server.connect(transport);
2215 | ```
2216 | </Tab>
2217 |
2218 | <Tab title="TypeScript (Client)">
2219 | ```typescript
2220 | const client = new Client({
2221 | name: "example-client",
2222 | version: "1.0.0"
2223 | }, {
2224 | capabilities: {}
2225 | });
2226 |
2227 | const transport = new StdioClientTransport({
2228 | command: "./server",
2229 | args: ["--option", "value"]
2230 | });
2231 | await client.connect(transport);
2232 | ```
2233 | </Tab>
2234 |
2235 | <Tab title="Python (Server)">
2236 | ```python
2237 | app = Server("example-server")
2238 |
2239 | async with stdio_server() as streams:
2240 | await app.run(
2241 | streams[0],
2242 | streams[1],
2243 | app.create_initialization_options()
2244 | )
2245 | ```
2246 | </Tab>
2247 |
2248 | <Tab title="Python (Client)">
2249 | ```python
2250 | params = StdioServerParameters(
2251 | command="./server",
2252 | args=["--option", "value"]
2253 | )
2254 |
2255 | async with stdio_client(params) as streams:
2256 | async with ClientSession(streams[0], streams[1]) as session:
2257 | await session.initialize()
2258 | ```
2259 | </Tab>
2260 | </Tabs>
2261 |
2262 | ### Server-Sent Events (SSE)
2263 |
2264 | SSE transport enables server-to-client streaming with HTTP POST requests for client-to-server communication.
2265 |
2266 | Use SSE when:
2267 |
2268 | * Only server-to-client streaming is needed
2269 | * Working with restricted networks
2270 | * Implementing simple updates
2271 |
2272 | <Tabs>
2273 | <Tab title="TypeScript (Server)">
2274 | ```typescript
2275 | import express from "express";
2276 |
2277 | const app = express();
2278 |
2279 | const server = new Server({
2280 | name: "example-server",
2281 | version: "1.0.0"
2282 | }, {
2283 | capabilities: {}
2284 | });
2285 |
2286 | let transport: SSEServerTransport | null = null;
2287 |
2288 | app.get("/sse", (req, res) => {
2289 | transport = new SSEServerTransport("/messages", res);
2290 | server.connect(transport);
2291 | });
2292 |
2293 | app.post("/messages", (req, res) => {
2294 | if (transport) {
2295 | transport.handlePostMessage(req, res);
2296 | }
2297 | });
2298 |
2299 | app.listen(3000);
2300 | ```
2301 | </Tab>
2302 |
2303 | <Tab title="TypeScript (Client)">
2304 | ```typescript
2305 | const client = new Client({
2306 | name: "example-client",
2307 | version: "1.0.0"
2308 | }, {
2309 | capabilities: {}
2310 | });
2311 |
2312 | const transport = new SSEClientTransport(
2313 | new URL("http://localhost:3000/sse")
2314 | );
2315 | await client.connect(transport);
2316 | ```
2317 | </Tab>
2318 |
2319 | <Tab title="Python (Server)">
2320 | ```python
2321 | from mcp.server.sse import SseServerTransport
2322 | from starlette.applications import Starlette
2323 | from starlette.routing import Route
2324 |
2325 | app = Server("example-server")
2326 | sse = SseServerTransport("/messages")
2327 |
2328 | async def handle_sse(scope, receive, send):
2329 | async with sse.connect_sse(scope, receive, send) as streams:
2330 | await app.run(streams[0], streams[1], app.create_initialization_options())
2331 |
2332 | async def handle_messages(scope, receive, send):
2333 | await sse.handle_post_message(scope, receive, send)
2334 |
2335 | starlette_app = Starlette(
2336 | routes=[
2337 | Route("/sse", endpoint=handle_sse),
2338 | Route("/messages", endpoint=handle_messages, methods=["POST"]),
2339 | ]
2340 | )
2341 | ```
2342 | </Tab>
2343 |
2344 | <Tab title="Python (Client)">
2345 | ```python
2346 | async with sse_client("http://localhost:8000/sse") as streams:
2347 | async with ClientSession(streams[0], streams[1]) as session:
2348 | await session.initialize()
2349 | ```
2350 | </Tab>
2351 | </Tabs>
2352 |
2353 | ## Custom Transports
2354 |
2355 | MCP makes it easy to implement custom transports for specific needs. Any transport implementation just needs to conform to the Transport interface:
2356 |
2357 | You can implement custom transports for:
2358 |
2359 | * Custom network protocols
2360 | * Specialized communication channels
2361 | * Integration with existing systems
2362 | * Performance optimization
2363 |
2364 | <Tabs>
2365 | <Tab title="TypeScript">
2366 | ```typescript
2367 | interface Transport {
2368 | // Start processing messages
2369 | start(): Promise<void>;
2370 |
2371 | // Send a JSON-RPC message
2372 | send(message: JSONRPCMessage): Promise<void>;
2373 |
2374 | // Close the connection
2375 | close(): Promise<void>;
2376 |
2377 | // Callbacks
2378 | onclose?: () => void;
2379 | onerror?: (error: Error) => void;
2380 | onmessage?: (message: JSONRPCMessage) => void;
2381 | }
2382 | ```
2383 | </Tab>
2384 |
2385 | <Tab title="Python">
2386 | Note that while MCP Servers are often implemented with asyncio, we recommend
2387 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2388 |
2389 | ```python
2390 | @contextmanager
2391 | async def create_transport(
2392 | read_stream: MemoryObjectReceiveStream[JSONRPCMessage | Exception],
2393 | write_stream: MemoryObjectSendStream[JSONRPCMessage]
2394 | ):
2395 | """
2396 | Transport interface for MCP.
2397 |
2398 | Args:
2399 | read_stream: Stream to read incoming messages from
2400 | write_stream: Stream to write outgoing messages to
2401 | """
2402 | async with anyio.create_task_group() as tg:
2403 | try:
2404 | # Start processing messages
2405 | tg.start_soon(lambda: process_messages(read_stream))
2406 |
2407 | # Send messages
2408 | async with write_stream:
2409 | yield write_stream
2410 |
2411 | except Exception as exc:
2412 | # Handle errors
2413 | raise exc
2414 | finally:
2415 | # Clean up
2416 | tg.cancel_scope.cancel()
2417 | await write_stream.aclose()
2418 | await read_stream.aclose()
2419 | ```
2420 | </Tab>
2421 | </Tabs>
2422 |
2423 | ## Error Handling
2424 |
2425 | Transport implementations should handle various error scenarios:
2426 |
2427 | 1. Connection errors
2428 | 2. Message parsing errors
2429 | 3. Protocol errors
2430 | 4. Network timeouts
2431 | 5. Resource cleanup
2432 |
2433 | Example error handling:
2434 |
2435 | <Tabs>
2436 | <Tab title="TypeScript">
2437 | ```typescript
2438 | class ExampleTransport implements Transport {
2439 | async start() {
2440 | try {
2441 | // Connection logic
2442 | } catch (error) {
2443 | this.onerror?.(new Error(`Failed to connect: ${error}`));
2444 | throw error;
2445 | }
2446 | }
2447 |
2448 | async send(message: JSONRPCMessage) {
2449 | try {
2450 | // Sending logic
2451 | } catch (error) {
2452 | this.onerror?.(new Error(`Failed to send message: ${error}`));
2453 | throw error;
2454 | }
2455 | }
2456 | }
2457 | ```
2458 | </Tab>
2459 |
2460 | <Tab title="Python">
2461 | Note that while MCP Servers are often implemented with asyncio, we recommend
2462 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2463 |
2464 | ```python
2465 | @contextmanager
2466 | async def example_transport(scope: Scope, receive: Receive, send: Send):
2467 | try:
2468 | # Create streams for bidirectional communication
2469 | read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
2470 | write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
2471 |
2472 | async def message_handler():
2473 | try:
2474 | async with read_stream_writer:
2475 | # Message handling logic
2476 | pass
2477 | except Exception as exc:
2478 | logger.error(f"Failed to handle message: {exc}")
2479 | raise exc
2480 |
2481 | async with anyio.create_task_group() as tg:
2482 | tg.start_soon(message_handler)
2483 | try:
2484 | # Yield streams for communication
2485 | yield read_stream, write_stream
2486 | except Exception as exc:
2487 | logger.error(f"Transport error: {exc}")
2488 | raise exc
2489 | finally:
2490 | tg.cancel_scope.cancel()
2491 | await write_stream.aclose()
2492 | await read_stream.aclose()
2493 | except Exception as exc:
2494 | logger.error(f"Failed to initialize transport: {exc}")
2495 | raise exc
2496 | ```
2497 | </Tab>
2498 | </Tabs>
2499 |
2500 | ## Best Practices
2501 |
2502 | When implementing or using MCP transport:
2503 |
2504 | 1. Handle connection lifecycle properly
2505 | 2. Implement proper error handling
2506 | 3. Clean up resources on connection close
2507 | 4. Use appropriate timeouts
2508 | 5. Validate messages before sending
2509 | 6. Log transport events for debugging
2510 | 7. Implement reconnection logic when appropriate
2511 | 8. Handle backpressure in message queues
2512 | 9. Monitor connection health
2513 | 10. Implement proper security measures
2514 |
2515 | ## Security Considerations
2516 |
2517 | When implementing transport:
2518 |
2519 | ### Authentication and Authorization
2520 |
2521 | * Implement proper authentication mechanisms
2522 | * Validate client credentials
2523 | * Use secure token handling
2524 | * Implement authorization checks
2525 |
2526 | ### Data Security
2527 |
2528 | * Use TLS for network transport
2529 | * Encrypt sensitive data
2530 | * Validate message integrity
2531 | * Implement message size limits
2532 | * Sanitize input data
2533 |
2534 | ### Network Security
2535 |
2536 | * Implement rate limiting
2537 | * Use appropriate timeouts
2538 | * Handle denial of service scenarios
2539 | * Monitor for unusual patterns
2540 | * Implement proper firewall rules
2541 |
2542 | ## Debugging Transport
2543 |
2544 | Tips for debugging transport issues:
2545 |
2546 | 1. Enable debug logging
2547 | 2. Monitor message flow
2548 | 3. Check connection states
2549 | 4. Validate message formats
2550 | 5. Test error scenarios
2551 | 6. Use network analysis tools
2552 | 7. Implement health checks
2553 | 8. Monitor resource usage
2554 | 9. Test edge cases
2555 | 10. Use proper error tracking
2556 |
2557 |
2558 | # Debugging
2559 | Source: https://modelcontextprotocol.io/docs/tools/debugging
2560 |
2561 | A comprehensive guide to debugging Model Context Protocol (MCP) integrations
2562 |
2563 | Effective debugging is essential when developing MCP servers or integrating them with applications. This guide covers the debugging tools and approaches available in the MCP ecosystem.
2564 |
2565 | <Info>
2566 | This guide is for macOS. Guides for other platforms are coming soon.
2567 | </Info>
2568 |
2569 | ## Debugging tools overview
2570 |
2571 | MCP provides several tools for debugging at different levels:
2572 |
2573 | 1. **MCP Inspector**
2574 | * Interactive debugging interface
2575 | * Direct server testing
2576 | * See the [Inspector guide](/docs/tools/inspector) for details
2577 |
2578 | 2. **Claude Desktop Developer Tools**
2579 | * Integration testing
2580 | * Log collection
2581 | * Chrome DevTools integration
2582 |
2583 | 3. **Server Logging**
2584 | * Custom logging implementations
2585 | * Error tracking
2586 | * Performance monitoring
2587 |
2588 | ## Debugging in Claude Desktop
2589 |
2590 | ### Checking server status
2591 |
2592 | The Claude.app interface provides basic server status information:
2593 |
2594 | 1. Click the <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/claude-desktop-mcp-plug-icon.svg" style={{display: 'inline', margin: 0, height: '1.3em'}} /> icon to view:
2595 | * Connected servers
2596 | * Available prompts and resources
2597 |
2598 | 2. Click the <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/claude-desktop-mcp-hammer-icon.svg" style={{display: 'inline', margin: 0, height: '1.3em'}} /> icon to view:
2599 | * Tools made available to the model
2600 |
2601 | ### Viewing logs
2602 |
2603 | Review detailed MCP logs from Claude Desktop:
2604 |
2605 | ```bash
2606 | # Follow logs in real-time
2607 | tail -n 20 -F ~/Library/Logs/Claude/mcp*.log
2608 | ```
2609 |
2610 | The logs capture:
2611 |
2612 | * Server connection events
2613 | * Configuration issues
2614 | * Runtime errors
2615 | * Message exchanges
2616 |
2617 | ### Using Chrome DevTools
2618 |
2619 | Access Chrome's developer tools inside Claude Desktop to investigate client-side errors:
2620 |
2621 | 1. Create a `developer_settings.json` file with `allowDevTools` set to true:
2622 |
2623 | ```bash
2624 | echo '{"allowDevTools": true}' > ~/Library/Application\ Support/Claude/developer_settings.json
2625 | ```
2626 |
2627 | 2. Open DevTools: `Command-Option-Shift-i`
2628 |
2629 | Note: You'll see two DevTools windows:
2630 |
2631 | * Main content window
2632 | * App title bar window
2633 |
2634 | Use the Console panel to inspect client-side errors.
2635 |
2636 | Use the Network panel to inspect:
2637 |
2638 | * Message payloads
2639 | * Connection timing
2640 |
2641 | ## Common issues
2642 |
2643 | ### Working directory
2644 |
2645 | When using MCP servers with Claude Desktop:
2646 |
2647 | * The working directory for servers launched via `claude_desktop_config.json` may be undefined (like `/` on macOS) since Claude Desktop could be started from anywhere
2648 | * Always use absolute paths in your configuration and `.env` files to ensure reliable operation
2649 | * For testing servers directly via command line, the working directory will be where you run the command
2650 |
2651 | For example in `claude_desktop_config.json`, use:
2652 |
2653 | ```json
2654 | {
2655 | "command": "npx",
2656 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/username/data"]
2657 | }
2658 | ```
2659 |
2660 | Instead of relative paths like `./data`
2661 |
2662 | ### Environment variables
2663 |
2664 | MCP servers inherit only a subset of environment variables automatically, like `USER`, `HOME`, and `PATH`.
2665 |
2666 | To override the default variables or provide your own, you can specify an `env` key in `claude_desktop_config.json`:
2667 |
2668 | ```json
2669 | {
2670 | "myserver": {
2671 | "command": "mcp-server-myapp",
2672 | "env": {
2673 | "MYAPP_API_KEY": "some_key",
2674 | }
2675 | }
2676 | }
2677 | ```
2678 |
2679 | ### Server initialization
2680 |
2681 | Common initialization problems:
2682 |
2683 | 1. **Path Issues**
2684 | * Incorrect server executable path
2685 | * Missing required files
2686 | * Permission problems
2687 | * Try using an absolute path for `command`
2688 |
2689 | 2. **Configuration Errors**
2690 | * Invalid JSON syntax
2691 | * Missing required fields
2692 | * Type mismatches
2693 |
2694 | 3. **Environment Problems**
2695 | * Missing environment variables
2696 | * Incorrect variable values
2697 | * Permission restrictions
2698 |
2699 | ### Connection problems
2700 |
2701 | When servers fail to connect:
2702 |
2703 | 1. Check Claude Desktop logs
2704 | 2. Verify server process is running
2705 | 3. Test standalone with [Inspector](/docs/tools/inspector)
2706 | 4. Verify protocol compatibility
2707 |
2708 | ## Implementing logging
2709 |
2710 | ### Server-side logging
2711 |
2712 | When building a server that uses the local stdio [transport](/docs/concepts/transports), all messages logged to stderr (standard error) will be captured by the host application (e.g., Claude Desktop) automatically.
2713 |
2714 | <Warning>
2715 | Local MCP servers should not log messages to stdout (standard out), as this will interfere with protocol operation.
2716 | </Warning>
2717 |
2718 | For all [transports](/docs/concepts/transports), you can also provide logging to the client by sending a log message notification:
2719 |
2720 | <Tabs>
2721 | <Tab title="Python">
2722 | ```python
2723 | server.request_context.session.send_log_message(
2724 | level="info",
2725 | data="Server started successfully",
2726 | )
2727 | ```
2728 | </Tab>
2729 |
2730 | <Tab title="TypeScript">
2731 | ```typescript
2732 | server.sendLoggingMessage({
2733 | level: "info",
2734 | data: "Server started successfully",
2735 | });
2736 | ```
2737 | </Tab>
2738 | </Tabs>
2739 |
2740 | Important events to log:
2741 |
2742 | * Initialization steps
2743 | * Resource access
2744 | * Tool execution
2745 | * Error conditions
2746 | * Performance metrics
2747 |
2748 | ### Client-side logging
2749 |
2750 | In client applications:
2751 |
2752 | 1. Enable debug logging
2753 | 2. Monitor network traffic
2754 | 3. Track message exchanges
2755 | 4. Record error states
2756 |
2757 | ## Debugging workflow
2758 |
2759 | ### Development cycle
2760 |
2761 | 1. Initial Development
2762 | * Use [Inspector](/docs/tools/inspector) for basic testing
2763 | * Implement core functionality
2764 | * Add logging points
2765 |
2766 | 2. Integration Testing
2767 | * Test in Claude Desktop
2768 | * Monitor logs
2769 | * Check error handling
2770 |
2771 | ### Testing changes
2772 |
2773 | To test changes efficiently:
2774 |
2775 | * **Configuration changes**: Restart Claude Desktop
2776 | * **Server code changes**: Use Command-R to reload
2777 | * **Quick iteration**: Use [Inspector](/docs/tools/inspector) during development
2778 |
2779 | ## Best practices
2780 |
2781 | ### Logging strategy
2782 |
2783 | 1. **Structured Logging**
2784 | * Use consistent formats
2785 | * Include context
2786 | * Add timestamps
2787 | * Track request IDs
2788 |
2789 | 2. **Error Handling**
2790 | * Log stack traces
2791 | * Include error context
2792 | * Track error patterns
2793 | * Monitor recovery
2794 |
2795 | 3. **Performance Tracking**
2796 | * Log operation timing
2797 | * Monitor resource usage
2798 | * Track message sizes
2799 | * Measure latency
2800 |
2801 | ### Security considerations
2802 |
2803 | When debugging:
2804 |
2805 | 1. **Sensitive Data**
2806 | * Sanitize logs
2807 | * Protect credentials
2808 | * Mask personal information
2809 |
2810 | 2. **Access Control**
2811 | * Verify permissions
2812 | * Check authentication
2813 | * Monitor access patterns
2814 |
2815 | ## Getting help
2816 |
2817 | When encountering issues:
2818 |
2819 | 1. **First Steps**
2820 | * Check server logs
2821 | * Test with [Inspector](/docs/tools/inspector)
2822 | * Review configuration
2823 | * Verify environment
2824 |
2825 | 2. **Support Channels**
2826 | * GitHub issues
2827 | * GitHub discussions
2828 |
2829 | 3. **Providing Information**
2830 | * Log excerpts
2831 | * Configuration files
2832 | * Steps to reproduce
2833 | * Environment details
2834 |
2835 | ## Next steps
2836 |
2837 | <CardGroup cols={2}>
2838 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
2839 | Learn to use the MCP Inspector
2840 | </Card>
2841 | </CardGroup>
2842 |
2843 |
2844 | # Inspector
2845 | Source: https://modelcontextprotocol.io/docs/tools/inspector
2846 |
2847 | In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers
2848 |
2849 | The [MCP Inspector](https://github.com/modelcontextprotocol/inspector) is an interactive developer tool for testing and debugging MCP servers. While the [Debugging Guide](/docs/tools/debugging) covers the Inspector as part of the overall debugging toolkit, this document provides a detailed exploration of the Inspector's features and capabilities.
2850 |
2851 | ## Getting started
2852 |
2853 | ### Installation and basic usage
2854 |
2855 | The Inspector runs directly through `npx` without requiring installation:
2856 |
2857 | ```bash
2858 | npx @modelcontextprotocol/inspector <command>
2859 | ```
2860 |
2861 | ```bash
2862 | npx @modelcontextprotocol/inspector <command> <arg1> <arg2>
2863 | ```
2864 |
2865 | #### Inspecting servers from NPM or PyPi
2866 |
2867 | A common way to start server packages from [NPM](https://npmjs.com) or [PyPi](https://pypi.com).
2868 |
2869 | <Tabs>
2870 | <Tab title="NPM package">
2871 | ```bash
2872 | npx -y @modelcontextprotocol/inspector npx <package-name> <args>
2873 | # For example
2874 | npx -y @modelcontextprotocol/inspector npx server-postgres postgres://127.0.0.1/testdb
2875 | ```
2876 | </Tab>
2877 |
2878 | <Tab title="PyPi package">
2879 | ```bash
2880 | npx @modelcontextprotocol/inspector uvx <package-name> <args>
2881 | # For example
2882 | npx @modelcontextprotocol/inspector uvx mcp-server-git --repository ~/code/mcp/servers.git
2883 | ```
2884 | </Tab>
2885 | </Tabs>
2886 |
2887 | #### Inspecting locally developed servers
2888 |
2889 | To inspect servers locally developed or downloaded as a repository, the most common
2890 | way is:
2891 |
2892 | <Tabs>
2893 | <Tab title="TypeScript">
2894 | ```bash
2895 | npx @modelcontextprotocol/inspector node path/to/server/index.js args...
2896 | ```
2897 | </Tab>
2898 |
2899 | <Tab title="Python">
2900 | ```bash
2901 | npx @modelcontextprotocol/inspector \
2902 | uv \
2903 | --directory path/to/server \
2904 | run \
2905 | package-name \
2906 | args...
2907 | ```
2908 | </Tab>
2909 | </Tabs>
2910 |
2911 | Please carefully read any attached README for the most accurate instructions.
2912 |
2913 | ## Feature overview
2914 |
2915 | <Frame caption="The MCP Inspector interface">
2916 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/mcp-inspector.png" />
2917 | </Frame>
2918 |
2919 | The Inspector provides several features for interacting with your MCP server:
2920 |
2921 | ### Server connection pane
2922 |
2923 | * Allows selecting the [transport](/docs/concepts/transports) for connecting to the server
2924 | * For local servers, supports customizing the command-line arguments and environment
2925 |
2926 | ### Resources tab
2927 |
2928 | * Lists all available resources
2929 | * Shows resource metadata (MIME types, descriptions)
2930 | * Allows resource content inspection
2931 | * Supports subscription testing
2932 |
2933 | ### Prompts tab
2934 |
2935 | * Displays available prompt templates
2936 | * Shows prompt arguments and descriptions
2937 | * Enables prompt testing with custom arguments
2938 | * Previews generated messages
2939 |
2940 | ### Tools tab
2941 |
2942 | * Lists available tools
2943 | * Shows tool schemas and descriptions
2944 | * Enables tool testing with custom inputs
2945 | * Displays tool execution results
2946 |
2947 | ### Notifications pane
2948 |
2949 | * Presents all logs recorded from the server
2950 | * Shows notifications received from the server
2951 |
2952 | ## Best practices
2953 |
2954 | ### Development workflow
2955 |
2956 | 1. Start Development
2957 | * Launch Inspector with your server
2958 | * Verify basic connectivity
2959 | * Check capability negotiation
2960 |
2961 | 2. Iterative testing
2962 | * Make server changes
2963 | * Rebuild the server
2964 | * Reconnect the Inspector
2965 | * Test affected features
2966 | * Monitor messages
2967 |
2968 | 3. Test edge cases
2969 | * Invalid inputs
2970 | * Missing prompt arguments
2971 | * Concurrent operations
2972 | * Verify error handling and error responses
2973 |
2974 | ## Next steps
2975 |
2976 | <CardGroup cols={2}>
2977 | <Card title="Inspector Repository" icon="github" href="https://github.com/modelcontextprotocol/inspector">
2978 | Check out the MCP Inspector source code
2979 | </Card>
2980 |
2981 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
2982 | Learn about broader debugging strategies
2983 | </Card>
2984 | </CardGroup>
2985 |
2986 |
2987 | # Example Servers
2988 | Source: https://modelcontextprotocol.io/examples
2989 |
2990 | A list of example servers and implementations
2991 |
2992 | This page showcases various Model Context Protocol (MCP) servers that demonstrate the protocol's capabilities and versatility. These servers enable Large Language Models (LLMs) to securely access tools and data sources.
2993 |
2994 | ## Reference implementations
2995 |
2996 | These official reference servers demonstrate core MCP features and SDK usage:
2997 |
2998 | ### Data and file systems
2999 |
3000 | * **[Filesystem](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)** - Secure file operations with configurable access controls
3001 | * **[PostgreSQL](https://github.com/modelcontextprotocol/servers/tree/main/src/postgres)** - Read-only database access with schema inspection capabilities
3002 | * **[SQLite](https://github.com/modelcontextprotocol/servers/tree/main/src/sqlite)** - Database interaction and business intelligence features
3003 | * **[Google Drive](https://github.com/modelcontextprotocol/servers/tree/main/src/gdrive)** - File access and search capabilities for Google Drive
3004 |
3005 | ### Development tools
3006 |
3007 | * **[Git](https://github.com/modelcontextprotocol/servers/tree/main/src/git)** - Tools to read, search, and manipulate Git repositories
3008 | * **[GitHub](https://github.com/modelcontextprotocol/servers/tree/main/src/github)** - Repository management, file operations, and GitHub API integration
3009 | * **[GitLab](https://github.com/modelcontextprotocol/servers/tree/main/src/gitlab)** - GitLab API integration enabling project management
3010 | * **[Sentry](https://github.com/modelcontextprotocol/servers/tree/main/src/sentry)** - Retrieving and analyzing issues from Sentry.io
3011 |
3012 | ### Web and browser automation
3013 |
3014 | * **[Brave Search](https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search)** - Web and local search using Brave's Search API
3015 | * **[Fetch](https://github.com/modelcontextprotocol/servers/tree/main/src/fetch)** - Web content fetching and conversion optimized for LLM usage
3016 | * **[Puppeteer](https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer)** - Browser automation and web scraping capabilities
3017 |
3018 | ### Productivity and communication
3019 |
3020 | * **[Slack](https://github.com/modelcontextprotocol/servers/tree/main/src/slack)** - Channel management and messaging capabilities
3021 | * **[Google Maps](https://github.com/modelcontextprotocol/servers/tree/main/src/google-maps)** - Location services, directions, and place details
3022 | * **[Memory](https://github.com/modelcontextprotocol/servers/tree/main/src/memory)** - Knowledge graph-based persistent memory system
3023 |
3024 | ### AI and specialized tools
3025 |
3026 | * **[EverArt](https://github.com/modelcontextprotocol/servers/tree/main/src/everart)** - AI image generation using various models
3027 | * **[Sequential Thinking](https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking)** - Dynamic problem-solving through thought sequences
3028 | * **[AWS KB Retrieval](https://github.com/modelcontextprotocol/servers/tree/main/src/aws-kb-retrieval-server)** - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime
3029 |
3030 | ## Official integrations
3031 |
3032 | These MCP servers are maintained by companies for their platforms:
3033 |
3034 | * **[Axiom](https://github.com/axiomhq/mcp-server-axiom)** - Query and analyze logs, traces, and event data using natural language
3035 | * **[Browserbase](https://github.com/browserbase/mcp-server-browserbase)** - Automate browser interactions in the cloud
3036 | * **[Cloudflare](https://github.com/cloudflare/mcp-server-cloudflare)** - Deploy and manage resources on the Cloudflare developer platform
3037 | * **[E2B](https://github.com/e2b-dev/mcp-server)** - Execute code in secure cloud sandboxes
3038 | * **[Neon](https://github.com/neondatabase/mcp-server-neon)** - Interact with the Neon serverless Postgres platform
3039 | * **[Obsidian Markdown Notes](https://github.com/calclavia/mcp-obsidian)** - Read and search through Markdown notes in Obsidian vaults
3040 | * **[Qdrant](https://github.com/qdrant/mcp-server-qdrant/)** - Implement semantic memory using the Qdrant vector search engine
3041 | * **[Raygun](https://github.com/MindscapeHQ/mcp-server-raygun)** - Access crash reporting and monitoring data
3042 | * **[Search1API](https://github.com/fatwang2/search1api-mcp)** - Unified API for search, crawling, and sitemaps
3043 | * **[Stripe](https://github.com/stripe/agent-toolkit)** - Interact with the Stripe API
3044 | * **[Tinybird](https://github.com/tinybirdco/mcp-tinybird)** - Interface with the Tinybird serverless ClickHouse platform
3045 | * **[Weaviate](https://github.com/weaviate/mcp-server-weaviate)** - Enable Agentic RAG through your Weaviate collection(s)
3046 |
3047 | ## Community highlights
3048 |
3049 | A growing ecosystem of community-developed servers extends MCP's capabilities:
3050 |
3051 | * **[Docker](https://github.com/ckreiling/mcp-server-docker)** - Manage containers, images, volumes, and networks
3052 | * **[Kubernetes](https://github.com/Flux159/mcp-server-kubernetes)** - Manage pods, deployments, and services
3053 | * **[Linear](https://github.com/jerhadf/linear-mcp-server)** - Project management and issue tracking
3054 | * **[Snowflake](https://github.com/datawiz168/mcp-snowflake-service)** - Interact with Snowflake databases
3055 | * **[Spotify](https://github.com/varunneal/spotify-mcp)** - Control Spotify playback and manage playlists
3056 | * **[Todoist](https://github.com/abhiz123/todoist-mcp-server)** - Task management integration
3057 |
3058 | > **Note:** Community servers are untested and should be used at your own risk. They are not affiliated with or endorsed by Anthropic.
3059 |
3060 | For a complete list of community servers, visit the [MCP Servers Repository](https://github.com/modelcontextprotocol/servers).
3061 |
3062 | ## Getting started
3063 |
3064 | ### Using reference servers
3065 |
3066 | TypeScript-based servers can be used directly with `npx`:
3067 |
3068 | ```bash
3069 | npx -y @modelcontextprotocol/server-memory
3070 | ```
3071 |
3072 | Python-based servers can be used with `uvx` (recommended) or `pip`:
3073 |
3074 | ```bash
3075 | # Using uvx
3076 | uvx mcp-server-git
3077 |
3078 | # Using pip
3079 | pip install mcp-server-git
3080 | python -m mcp_server_git
3081 | ```
3082 |
3083 | ### Configuring with Claude
3084 |
3085 | To use an MCP server with Claude, add it to your configuration:
3086 |
3087 | ```json
3088 | {
3089 | "mcpServers": {
3090 | "memory": {
3091 | "command": "npx",
3092 | "args": ["-y", "@modelcontextprotocol/server-memory"]
3093 | },
3094 | "filesystem": {
3095 | "command": "npx",
3096 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
3097 | },
3098 | "github": {
3099 | "command": "npx",
3100 | "args": ["-y", "@modelcontextprotocol/server-github"],
3101 | "env": {
3102 | "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
3103 | }
3104 | }
3105 | }
3106 | }
3107 | ```
3108 |
3109 | ## Additional resources
3110 |
3111 | * [MCP Servers Repository](https://github.com/modelcontextprotocol/servers) - Complete collection of reference implementations and community servers
3112 | * [Awesome MCP Servers](https://github.com/punkpeye/awesome-mcp-servers) - Curated list of MCP servers
3113 | * [MCP CLI](https://github.com/wong2/mcp-cli) - Command-line inspector for testing MCP servers
3114 | * [MCP Get](https://mcp-get.com) - Tool for installing and managing MCP servers
3115 | * [Supergateway](https://github.com/supercorp-ai/supergateway) - Run MCP stdio servers over SSE
3116 |
3117 | Visit our [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions) to engage with the MCP community.
3118 |
3119 |
3120 | # Introduction
3121 | Source: https://modelcontextprotocol.io/introduction
3122 |
3123 | Get started with the Model Context Protocol (MCP)
3124 |
3125 | <Note>Java SDK released! Check out [what else is new.](/development/updates)</Note>
3126 |
3127 | MCP is an open protocol that standardizes how applications provide context to LLMs. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.
3128 |
3129 | ## Why MCP?
3130 |
3131 | MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:
3132 |
3133 | * A growing list of pre-built integrations that your LLM can directly plug into
3134 | * The flexibility to switch between LLM providers and vendors
3135 | * Best practices for securing your data within your infrastructure
3136 |
3137 | ### General architecture
3138 |
3139 | At its core, MCP follows a client-server architecture where a host application can connect to multiple servers:
3140 |
3141 | ```mermaid
3142 | flowchart LR
3143 | subgraph "Your Computer"
3144 | Host["Host with MCP Client\n(Claude, IDEs, Tools)"]
3145 | S1["MCP Server A"]
3146 | S2["MCP Server B"]
3147 | S3["MCP Server C"]
3148 | Host <-->|"MCP Protocol"| S1
3149 | Host <-->|"MCP Protocol"| S2
3150 | Host <-->|"MCP Protocol"| S3
3151 | S1 <--> D1[("Local\nData Source A")]
3152 | S2 <--> D2[("Local\nData Source B")]
3153 | end
3154 | subgraph "Internet"
3155 | S3 <-->|"Web APIs"| D3[("Remote\nService C")]
3156 | end
3157 | ```
3158 |
3159 | * **MCP Hosts**: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
3160 | * **MCP Clients**: Protocol clients that maintain 1:1 connections with servers
3161 | * **MCP Servers**: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
3162 | * **Local Data Sources**: Your computer's files, databases, and services that MCP servers can securely access
3163 | * **Remote Services**: External systems available over the internet (e.g., through APIs) that MCP servers can connect to
3164 |
3165 | ## Get started
3166 |
3167 | Choose the path that best fits your needs:
3168 |
3169 | #### Quick Starts
3170 |
3171 | <CardGroup cols={2}>
3172 | <Card title="For Server Developers" icon="bolt" href="/quickstart/server">
3173 | Get started building your own server to use in Claude for Desktop and other clients
3174 | </Card>
3175 |
3176 | <Card title="For Client Developers" icon="bolt" href="/quickstart/client">
3177 | Get started building your own client that can integrate with all MCP servers
3178 | </Card>
3179 |
3180 | <Card title="For Claude Desktop Users" icon="bolt" href="/quickstart/user">
3181 | Get started using pre-built servers in Claude for Desktop
3182 | </Card>
3183 | </CardGroup>
3184 |
3185 | #### Examples
3186 |
3187 | <CardGroup cols={2}>
3188 | <Card title="Example Servers" icon="grid" href="/examples">
3189 | Check out our gallery of official MCP servers and implementations
3190 | </Card>
3191 |
3192 | <Card title="Example Clients" icon="cubes" href="/clients">
3193 | View the list of clients that support MCP integrations
3194 | </Card>
3195 | </CardGroup>
3196 |
3197 | ## Tutorials
3198 |
3199 | <CardGroup cols={2}>
3200 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
3201 | Learn how to use LLMs like Claude to speed up your MCP development
3202 | </Card>
3203 |
3204 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
3205 | Learn how to effectively debug MCP servers and integrations
3206 | </Card>
3207 |
3208 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
3209 | Test and inspect your MCP servers with our interactive debugging tool
3210 | </Card>
3211 |
3212 | <Card title="MCP Workshop (Video, 2hr)" icon="person-chalkboard" href="https://www.youtube.com/watch?v=kQmXtrmQ5Zg">
3213 | <iframe src="https://www.youtube.com/embed/kQmXtrmQ5Zg" />
3214 | </Card>
3215 | </CardGroup>
3216 |
3217 | ## Explore MCP
3218 |
3219 | Dive deeper into MCP's core concepts and capabilities:
3220 |
3221 | <CardGroup cols={2}>
3222 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3223 | Understand how MCP connects clients, servers, and LLMs
3224 | </Card>
3225 |
3226 | <Card title="Resources" icon="database" href="/docs/concepts/resources">
3227 | Expose data and content from your servers to LLMs
3228 | </Card>
3229 |
3230 | <Card title="Prompts" icon="message" href="/docs/concepts/prompts">
3231 | Create reusable prompt templates and workflows
3232 | </Card>
3233 |
3234 | <Card title="Tools" icon="wrench" href="/docs/concepts/tools">
3235 | Enable LLMs to perform actions through your server
3236 | </Card>
3237 |
3238 | <Card title="Sampling" icon="robot" href="/docs/concepts/sampling">
3239 | Let your servers request completions from LLMs
3240 | </Card>
3241 |
3242 | <Card title="Transports" icon="network-wired" href="/docs/concepts/transports">
3243 | Learn about MCP's communication mechanism
3244 | </Card>
3245 | </CardGroup>
3246 |
3247 | ## Contributing
3248 |
3249 | Want to contribute? Check out our [Contributing Guide](/development/contributing) to learn how you can help improve MCP.
3250 |
3251 | ## Support and Feedback
3252 |
3253 | Here's how to get help or provide feedback:
3254 |
3255 | * For bug reports and feature requests related to the MCP specification, SDKs, or documentation (open source), please [create a GitHub issue](https://github.com/modelcontextprotocol)
3256 | * For discussions or Q\&A about the MCP specification, use the [specification discussions](https://github.com/modelcontextprotocol/specification/discussions)
3257 | * For discussions or Q\&A about other MCP open source components, use the [organization discussions](https://github.com/orgs/modelcontextprotocol/discussions)
3258 | * For bug reports, feature requests, and questions related to Claude.app and claude.ai's MCP integration, please email [[email protected]](mailto:[email protected])
3259 |
3260 |
3261 | # For Client Developers
3262 | Source: https://modelcontextprotocol.io/quickstart/client
3263 |
3264 | Get started building your own client that can integrate with all MCP servers.
3265 |
3266 | In this tutorial, you'll learn how to build a LLM-powered chatbot client that connects to MCP servers. It helps to have gone through the [Server quickstart](/quickstart/server) that guides you through the basic of building your first server.
3267 |
3268 | <Tabs>
3269 | <Tab title="Python">
3270 | [You can find the complete code for this tutorial here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/mcp-client-python)
3271 |
3272 | ## System Requirements
3273 |
3274 | Before starting, ensure your system meets these requirements:
3275 |
3276 | * Mac or Windows computer
3277 | * Latest Python version installed
3278 | * Latest version of `uv` installed
3279 |
3280 | ## Setting Up Your Environment
3281 |
3282 | First, create a new Python project with `uv`:
3283 |
3284 | ```bash
3285 | # Create project directory
3286 | uv init mcp-client
3287 | cd mcp-client
3288 |
3289 | # Create virtual environment
3290 | uv venv
3291 |
3292 | # Activate virtual environment
3293 | # On Windows:
3294 | .venv\Scripts\activate
3295 | # On Unix or MacOS:
3296 | source .venv/bin/activate
3297 |
3298 | # Install required packages
3299 | uv add mcp anthropic python-dotenv
3300 |
3301 | # Remove boilerplate files
3302 | rm hello.py
3303 |
3304 | # Create our main file
3305 | touch client.py
3306 | ```
3307 |
3308 | ## Setting Up Your API Key
3309 |
3310 | You'll need an Anthropic API key from the [Anthropic Console](https://console.anthropic.com/settings/keys).
3311 |
3312 | Create a `.env` file to store it:
3313 |
3314 | ```bash
3315 | # Create .env file
3316 | touch .env
3317 | ```
3318 |
3319 | Add your key to the `.env` file:
3320 |
3321 | ```bash
3322 | ANTHROPIC_API_KEY=<your key here>
3323 | ```
3324 |
3325 | Add `.env` to your `.gitignore`:
3326 |
3327 | ```bash
3328 | echo ".env" >> .gitignore
3329 | ```
3330 |
3331 | <Warning>
3332 | Make sure you keep your `ANTHROPIC_API_KEY` secure!
3333 | </Warning>
3334 |
3335 | ## Creating the Client
3336 |
3337 | ### Basic Client Structure
3338 |
3339 | First, let's set up our imports and create the basic client class:
3340 |
3341 | ```python
3342 | import asyncio
3343 | from typing import Optional
3344 | from contextlib import AsyncExitStack
3345 |
3346 | from mcp import ClientSession, StdioServerParameters
3347 | from mcp.client.stdio import stdio_client
3348 |
3349 | from anthropic import Anthropic
3350 | from dotenv import load_dotenv
3351 |
3352 | load_dotenv() # load environment variables from .env
3353 |
3354 | class MCPClient:
3355 | def __init__(self):
3356 | # Initialize session and client objects
3357 | self.session: Optional[ClientSession] = None
3358 | self.exit_stack = AsyncExitStack()
3359 | self.anthropic = Anthropic()
3360 | # methods will go here
3361 | ```
3362 |
3363 | ### Server Connection Management
3364 |
3365 | Next, we'll implement the method to connect to an MCP server:
3366 |
3367 | ```python
3368 | async def connect_to_server(self, server_script_path: str):
3369 | """Connect to an MCP server
3370 |
3371 | Args:
3372 | server_script_path: Path to the server script (.py or .js)
3373 | """
3374 | is_python = server_script_path.endswith('.py')
3375 | is_js = server_script_path.endswith('.js')
3376 | if not (is_python or is_js):
3377 | raise ValueError("Server script must be a .py or .js file")
3378 |
3379 | command = "python" if is_python else "node"
3380 | server_params = StdioServerParameters(
3381 | command=command,
3382 | args=[server_script_path],
3383 | env=None
3384 | )
3385 |
3386 | stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
3387 | self.stdio, self.write = stdio_transport
3388 | self.session = await self.exit_stack.enter_async_context(ClientSession(self.stdio, self.write))
3389 |
3390 | await self.session.initialize()
3391 |
3392 | # List available tools
3393 | response = await self.session.list_tools()
3394 | tools = response.tools
3395 | print("\nConnected to server with tools:", [tool.name for tool in tools])
3396 | ```
3397 |
3398 | ### Query Processing Logic
3399 |
3400 | Now let's add the core functionality for processing queries and handling tool calls:
3401 |
3402 | ```python
3403 | async def process_query(self, query: str) -> str:
3404 | """Process a query using Claude and available tools"""
3405 | messages = [
3406 | {
3407 | "role": "user",
3408 | "content": query
3409 | }
3410 | ]
3411 |
3412 | response = await self.session.list_tools()
3413 | available_tools = [{
3414 | "name": tool.name,
3415 | "description": tool.description,
3416 | "input_schema": tool.inputSchema
3417 | } for tool in response.tools]
3418 |
3419 | # Initial Claude API call
3420 | response = self.anthropic.messages.create(
3421 | model="claude-3-5-sonnet-20241022",
3422 | max_tokens=1000,
3423 | messages=messages,
3424 | tools=available_tools
3425 | )
3426 |
3427 | # Process response and handle tool calls
3428 | final_text = []
3429 |
3430 | assistant_message_content = []
3431 | for content in response.content:
3432 | if content.type == 'text':
3433 | final_text.append(content.text)
3434 | assistant_message_content.append(content)
3435 | elif content.type == 'tool_use':
3436 | tool_name = content.name
3437 | tool_args = content.input
3438 |
3439 | # Execute tool call
3440 | result = await self.session.call_tool(tool_name, tool_args)
3441 | final_text.append(f"[Calling tool {tool_name} with args {tool_args}]")
3442 |
3443 | assistant_message_content.append(content)
3444 | messages.append({
3445 | "role": "assistant",
3446 | "content": assistant_message_content
3447 | })
3448 | messages.append({
3449 | "role": "user",
3450 | "content": [
3451 | {
3452 | "type": "tool_result",
3453 | "tool_use_id": content.id,
3454 | "content": result.content
3455 | }
3456 | ]
3457 | })
3458 |
3459 | # Get next response from Claude
3460 | response = self.anthropic.messages.create(
3461 | model="claude-3-5-sonnet-20241022",
3462 | max_tokens=1000,
3463 | messages=messages,
3464 | tools=available_tools
3465 | )
3466 |
3467 | final_text.append(response.content[0].text)
3468 |
3469 | return "\n".join(final_text)
3470 | ```
3471 |
3472 | ### Interactive Chat Interface
3473 |
3474 | Now we'll add the chat loop and cleanup functionality:
3475 |
3476 | ```python
3477 | async def chat_loop(self):
3478 | """Run an interactive chat loop"""
3479 | print("\nMCP Client Started!")
3480 | print("Type your queries or 'quit' to exit.")
3481 |
3482 | while True:
3483 | try:
3484 | query = input("\nQuery: ").strip()
3485 |
3486 | if query.lower() == 'quit':
3487 | break
3488 |
3489 | response = await self.process_query(query)
3490 | print("\n" + response)
3491 |
3492 | except Exception as e:
3493 | print(f"\nError: {str(e)}")
3494 |
3495 | async def cleanup(self):
3496 | """Clean up resources"""
3497 | await self.exit_stack.aclose()
3498 | ```
3499 |
3500 | ### Main Entry Point
3501 |
3502 | Finally, we'll add the main execution logic:
3503 |
3504 | ```python
3505 | async def main():
3506 | if len(sys.argv) < 2:
3507 | print("Usage: python client.py <path_to_server_script>")
3508 | sys.exit(1)
3509 |
3510 | client = MCPClient()
3511 | try:
3512 | await client.connect_to_server(sys.argv[1])
3513 | await client.chat_loop()
3514 | finally:
3515 | await client.cleanup()
3516 |
3517 | if __name__ == "__main__":
3518 | import sys
3519 | asyncio.run(main())
3520 | ```
3521 |
3522 | You can find the complete `client.py` file [here.](https://gist.github.com/zckly/f3f28ea731e096e53b39b47bf0a2d4b1)
3523 |
3524 | ## Key Components Explained
3525 |
3526 | ### 1. Client Initialization
3527 |
3528 | * The `MCPClient` class initializes with session management and API clients
3529 | * Uses `AsyncExitStack` for proper resource management
3530 | * Configures the Anthropic client for Claude interactions
3531 |
3532 | ### 2. Server Connection
3533 |
3534 | * Supports both Python and Node.js servers
3535 | * Validates server script type
3536 | * Sets up proper communication channels
3537 | * Initializes the session and lists available tools
3538 |
3539 | ### 3. Query Processing
3540 |
3541 | * Maintains conversation context
3542 | * Handles Claude's responses and tool calls
3543 | * Manages the message flow between Claude and tools
3544 | * Combines results into a coherent response
3545 |
3546 | ### 4. Interactive Interface
3547 |
3548 | * Provides a simple command-line interface
3549 | * Handles user input and displays responses
3550 | * Includes basic error handling
3551 | * Allows graceful exit
3552 |
3553 | ### 5. Resource Management
3554 |
3555 | * Proper cleanup of resources
3556 | * Error handling for connection issues
3557 | * Graceful shutdown procedures
3558 |
3559 | ## Common Customization Points
3560 |
3561 | 1. **Tool Handling**
3562 | * Modify `process_query()` to handle specific tool types
3563 | * Add custom error handling for tool calls
3564 | * Implement tool-specific response formatting
3565 |
3566 | 2. **Response Processing**
3567 | * Customize how tool results are formatted
3568 | * Add response filtering or transformation
3569 | * Implement custom logging
3570 |
3571 | 3. **User Interface**
3572 | * Add a GUI or web interface
3573 | * Implement rich console output
3574 | * Add command history or auto-completion
3575 |
3576 | ## Running the Client
3577 |
3578 | To run your client with any MCP server:
3579 |
3580 | ```bash
3581 | uv run client.py path/to/server.py # python server
3582 | uv run client.py path/to/build/index.js # node server
3583 | ```
3584 |
3585 | <Note>
3586 | If you're continuing the weather tutorial from the server quickstart, your command might look something like this: `python client.py .../weather/src/weather/server.py`
3587 | </Note>
3588 |
3589 | The client will:
3590 |
3591 | 1. Connect to the specified server
3592 | 2. List available tools
3593 | 3. Start an interactive chat session where you can:
3594 | * Enter queries
3595 | * See tool executions
3596 | * Get responses from Claude
3597 |
3598 | Here's an example of what it should look like if connected to the weather server from the server quickstart:
3599 |
3600 | <Frame>
3601 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/client-claude-cli-python.png" />
3602 | </Frame>
3603 |
3604 | ## How It Works
3605 |
3606 | When you submit a query:
3607 |
3608 | 1. The client gets the list of available tools from the server
3609 | 2. Your query is sent to Claude along with tool descriptions
3610 | 3. Claude decides which tools (if any) to use
3611 | 4. The client executes any requested tool calls through the server
3612 | 5. Results are sent back to Claude
3613 | 6. Claude provides a natural language response
3614 | 7. The response is displayed to you
3615 |
3616 | ## Best practices
3617 |
3618 | 1. **Error Handling**
3619 | * Always wrap tool calls in try-catch blocks
3620 | * Provide meaningful error messages
3621 | * Gracefully handle connection issues
3622 |
3623 | 2. **Resource Management**
3624 | * Use `AsyncExitStack` for proper cleanup
3625 | * Close connections when done
3626 | * Handle server disconnections
3627 |
3628 | 3. **Security**
3629 | * Store API keys securely in `.env`
3630 | * Validate server responses
3631 | * Be cautious with tool permissions
3632 |
3633 | ## Troubleshooting
3634 |
3635 | ### Server Path Issues
3636 |
3637 | * Double-check the path to your server script is correct
3638 | * Use the absolute path if the relative path isn't working
3639 | * For Windows users, make sure to use forward slashes (/) or escaped backslashes (\\) in the path
3640 | * Verify the server file has the correct extension (.py for Python or .js for Node.js)
3641 |
3642 | Example of correct path usage:
3643 |
3644 | ```bash
3645 | # Relative path
3646 | uv run client.py ./server/weather.py
3647 |
3648 | # Absolute path
3649 | uv run client.py /Users/username/projects/mcp-server/weather.py
3650 |
3651 | # Windows path (either format works)
3652 | uv run client.py C:/projects/mcp-server/weather.py
3653 | uv run client.py C:\\projects\\mcp-server\\weather.py
3654 | ```
3655 |
3656 | ### Response Timing
3657 |
3658 | * The first response might take up to 30 seconds to return
3659 | * This is normal and happens while:
3660 | * The server initializes
3661 | * Claude processes the query
3662 | * Tools are being executed
3663 | * Subsequent responses are typically faster
3664 | * Don't interrupt the process during this initial waiting period
3665 |
3666 | ### Common Error Messages
3667 |
3668 | If you see:
3669 |
3670 | * `FileNotFoundError`: Check your server path
3671 | * `Connection refused`: Ensure the server is running and the path is correct
3672 | * `Tool execution failed`: Verify the tool's required environment variables are set
3673 | * `Timeout error`: Consider increasing the timeout in your client configuration
3674 | </Tab>
3675 |
3676 | <Tab title="Node">
3677 | [You can find the complete code for this tutorial here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/mcp-client-typescript)
3678 |
3679 | ## System Requirements
3680 |
3681 | Before starting, ensure your system meets these requirements:
3682 |
3683 | * Mac or Windows computer
3684 | * Node.js 16 or higher installed
3685 | * Latest version of `npm` installed
3686 | * Anthropic API key (Claude)
3687 |
3688 | ## Setting Up Your Environment
3689 |
3690 | First, let's create and set up our project:
3691 |
3692 | <CodeGroup>
3693 | ```bash MacOS/Linux
3694 | # Create project directory
3695 | mkdir mcp-client-typescript
3696 | cd mcp-client-typescript
3697 |
3698 | # Initialize npm project
3699 | npm init -y
3700 |
3701 | # Install dependencies
3702 | npm install @anthropic-ai/sdk @modelcontextprotocol/sdk dotenv
3703 |
3704 | # Install dev dependencies
3705 | npm install -D @types/node typescript
3706 |
3707 | # Create source file
3708 | touch index.ts
3709 | ```
3710 |
3711 | ```powershell Windows
3712 | # Create project directory
3713 | md mcp-client-typescript
3714 | cd mcp-client-typescript
3715 |
3716 | # Initialize npm project
3717 | npm init -y
3718 |
3719 | # Install dependencies
3720 | npm install @anthropic-ai/sdk @modelcontextprotocol/sdk dotenv
3721 |
3722 | # Install dev dependencies
3723 | npm install -D @types/node typescript
3724 |
3725 | # Create source file
3726 | new-item index.ts
3727 | ```
3728 | </CodeGroup>
3729 |
3730 | Update your `package.json` to set `type: "module"` and a build script:
3731 |
3732 | ```json package.json
3733 | {
3734 | "type": "module",
3735 | "scripts": {
3736 | "build": "tsc && chmod 755 build/index.js"
3737 | }
3738 | }
3739 | ```
3740 |
3741 | Create a `tsconfig.json` in the root of your project:
3742 |
3743 | ```json tsconfig.json
3744 | {
3745 | "compilerOptions": {
3746 | "target": "ES2022",
3747 | "module": "Node16",
3748 | "moduleResolution": "Node16",
3749 | "outDir": "./build",
3750 | "rootDir": "./",
3751 | "strict": true,
3752 | "esModuleInterop": true,
3753 | "skipLibCheck": true,
3754 | "forceConsistentCasingInFileNames": true
3755 | },
3756 | "include": ["index.ts"],
3757 | "exclude": ["node_modules"]
3758 | }
3759 | ```
3760 |
3761 | ## Setting Up Your API Key
3762 |
3763 | You'll need an Anthropic API key from the [Anthropic Console](https://console.anthropic.com/settings/keys).
3764 |
3765 | Create a `.env` file to store it:
3766 |
3767 | ```bash
3768 | echo "ANTHROPIC_API_KEY=<your key here>" > .env
3769 | ```
3770 |
3771 | Add `.env` to your `.gitignore`:
3772 |
3773 | ```bash
3774 | echo ".env" >> .gitignore
3775 | ```
3776 |
3777 | <Warning>
3778 | Make sure you keep your `ANTHROPIC_API_KEY` secure!
3779 | </Warning>
3780 |
3781 | ## Creating the Client
3782 |
3783 | ### Basic Client Structure
3784 |
3785 | First, let's set up our imports and create the basic client class in `index.ts`:
3786 |
3787 | ```typescript
3788 | import { Anthropic } from "@anthropic-ai/sdk";
3789 | import {
3790 | MessageParam,
3791 | Tool,
3792 | } from "@anthropic-ai/sdk/resources/messages/messages.mjs";
3793 | import { Client } from "@modelcontextprotocol/sdk/client/index.js";
3794 | import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
3795 | import readline from "readline/promises";
3796 | import dotenv from "dotenv";
3797 |
3798 | dotenv.config();
3799 |
3800 | const ANTHROPIC_API_KEY = process.env.ANTHROPIC_API_KEY;
3801 | if (!ANTHROPIC_API_KEY) {
3802 | throw new Error("ANTHROPIC_API_KEY is not set");
3803 | }
3804 |
3805 | class MCPClient {
3806 | private mcp: Client;
3807 | private anthropic: Anthropic;
3808 | private transport: StdioClientTransport | null = null;
3809 | private tools: Tool[] = [];
3810 |
3811 | constructor() {
3812 | this.anthropic = new Anthropic({
3813 | apiKey: ANTHROPIC_API_KEY,
3814 | });
3815 | this.mcp = new Client({ name: "mcp-client-cli", version: "1.0.0" });
3816 | }
3817 | // methods will go here
3818 | }
3819 | ```
3820 |
3821 | ### Server Connection Management
3822 |
3823 | Next, we'll implement the method to connect to an MCP server:
3824 |
3825 | ```typescript
3826 | async connectToServer(serverScriptPath: string) {
3827 | try {
3828 | const isJs = serverScriptPath.endsWith(".js");
3829 | const isPy = serverScriptPath.endsWith(".py");
3830 | if (!isJs && !isPy) {
3831 | throw new Error("Server script must be a .js or .py file");
3832 | }
3833 | const command = isPy
3834 | ? process.platform === "win32"
3835 | ? "python"
3836 | : "python3"
3837 | : process.execPath;
3838 |
3839 | this.transport = new StdioClientTransport({
3840 | command,
3841 | args: [serverScriptPath],
3842 | });
3843 | this.mcp.connect(this.transport);
3844 |
3845 | const toolsResult = await this.mcp.listTools();
3846 | this.tools = toolsResult.tools.map((tool) => {
3847 | return {
3848 | name: tool.name,
3849 | description: tool.description,
3850 | input_schema: tool.inputSchema,
3851 | };
3852 | });
3853 | console.log(
3854 | "Connected to server with tools:",
3855 | this.tools.map(({ name }) => name)
3856 | );
3857 | } catch (e) {
3858 | console.log("Failed to connect to MCP server: ", e);
3859 | throw e;
3860 | }
3861 | }
3862 | ```
3863 |
3864 | ### Query Processing Logic
3865 |
3866 | Now let's add the core functionality for processing queries and handling tool calls:
3867 |
3868 | ```typescript
3869 | async processQuery(query: string) {
3870 | const messages: MessageParam[] = [
3871 | {
3872 | role: "user",
3873 | content: query,
3874 | },
3875 | ];
3876 |
3877 | const response = await this.anthropic.messages.create({
3878 | model: "claude-3-5-sonnet-20241022",
3879 | max_tokens: 1000,
3880 | messages,
3881 | tools: this.tools,
3882 | });
3883 |
3884 | const finalText = [];
3885 | const toolResults = [];
3886 |
3887 | for (const content of response.content) {
3888 | if (content.type === "text") {
3889 | finalText.push(content.text);
3890 | } else if (content.type === "tool_use") {
3891 | const toolName = content.name;
3892 | const toolArgs = content.input as { [x: string]: unknown } | undefined;
3893 |
3894 | const result = await this.mcp.callTool({
3895 | name: toolName,
3896 | arguments: toolArgs,
3897 | });
3898 | toolResults.push(result);
3899 | finalText.push(
3900 | `[Calling tool ${toolName} with args ${JSON.stringify(toolArgs)}]`
3901 | );
3902 |
3903 | messages.push({
3904 | role: "user",
3905 | content: result.content as string,
3906 | });
3907 |
3908 | const response = await this.anthropic.messages.create({
3909 | model: "claude-3-5-sonnet-20241022",
3910 | max_tokens: 1000,
3911 | messages,
3912 | });
3913 |
3914 | finalText.push(
3915 | response.content[0].type === "text" ? response.content[0].text : ""
3916 | );
3917 | }
3918 | }
3919 |
3920 | return finalText.join("\n");
3921 | }
3922 | ```
3923 |
3924 | ### Interactive Chat Interface
3925 |
3926 | Now we'll add the chat loop and cleanup functionality:
3927 |
3928 | ```typescript
3929 | async chatLoop() {
3930 | const rl = readline.createInterface({
3931 | input: process.stdin,
3932 | output: process.stdout,
3933 | });
3934 |
3935 | try {
3936 | console.log("\nMCP Client Started!");
3937 | console.log("Type your queries or 'quit' to exit.");
3938 |
3939 | while (true) {
3940 | const message = await rl.question("\nQuery: ");
3941 | if (message.toLowerCase() === "quit") {
3942 | break;
3943 | }
3944 | const response = await this.processQuery(message);
3945 | console.log("\n" + response);
3946 | }
3947 | } finally {
3948 | rl.close();
3949 | }
3950 | }
3951 |
3952 | async cleanup() {
3953 | await this.mcp.close();
3954 | }
3955 | ```
3956 |
3957 | ### Main Entry Point
3958 |
3959 | Finally, we'll add the main execution logic:
3960 |
3961 | ```typescript
3962 | async function main() {
3963 | if (process.argv.length < 3) {
3964 | console.log("Usage: node index.ts <path_to_server_script>");
3965 | return;
3966 | }
3967 | const mcpClient = new MCPClient();
3968 | try {
3969 | await mcpClient.connectToServer(process.argv[2]);
3970 | await mcpClient.chatLoop();
3971 | } finally {
3972 | await mcpClient.cleanup();
3973 | process.exit(0);
3974 | }
3975 | }
3976 |
3977 | main();
3978 | ```
3979 |
3980 | ## Running the Client
3981 |
3982 | To run your client with any MCP server:
3983 |
3984 | ```bash
3985 | # Build TypeScript
3986 | npm run build
3987 |
3988 | # Run the client
3989 | node build/index.js path/to/server.py # python server
3990 | node build/index.js path/to/build/index.js # node server
3991 | ```
3992 |
3993 | <Note>
3994 | If you're continuing the weather tutorial from the server quickstart, your command might look something like this: `node build/index.js .../quickstart-resources/weather-server-typescript/build/index.js`
3995 | </Note>
3996 |
3997 | **The client will:**
3998 |
3999 | 1. Connect to the specified server
4000 | 2. List available tools
4001 | 3. Start an interactive chat session where you can:
4002 | * Enter queries
4003 | * See tool executions
4004 | * Get responses from Claude
4005 |
4006 | ## How It Works
4007 |
4008 | When you submit a query:
4009 |
4010 | 1. The client gets the list of available tools from the server
4011 | 2. Your query is sent to Claude along with tool descriptions
4012 | 3. Claude decides which tools (if any) to use
4013 | 4. The client executes any requested tool calls through the server
4014 | 5. Results are sent back to Claude
4015 | 6. Claude provides a natural language response
4016 | 7. The response is displayed to you
4017 |
4018 | ## Best practices
4019 |
4020 | 1. **Error Handling**
4021 | * Use TypeScript's type system for better error detection
4022 | * Wrap tool calls in try-catch blocks
4023 | * Provide meaningful error messages
4024 | * Gracefully handle connection issues
4025 |
4026 | 2. **Security**
4027 | * Store API keys securely in `.env`
4028 | * Validate server responses
4029 | * Be cautious with tool permissions
4030 |
4031 | ## Troubleshooting
4032 |
4033 | ### Server Path Issues
4034 |
4035 | * Double-check the path to your server script is correct
4036 | * Use the absolute path if the relative path isn't working
4037 | * For Windows users, make sure to use forward slashes (/) or escaped backslashes (\\) in the path
4038 | * Verify the server file has the correct extension (.js for Node.js or .py for Python)
4039 |
4040 | Example of correct path usage:
4041 |
4042 | ```bash
4043 | # Relative path
4044 | node build/index.js ./server/build/index.js
4045 |
4046 | # Absolute path
4047 | node build/index.js /Users/username/projects/mcp-server/build/index.js
4048 |
4049 | # Windows path (either format works)
4050 | node build/index.js C:/projects/mcp-server/build/index.js
4051 | node build/index.js C:\\projects\\mcp-server\\build\\index.js
4052 | ```
4053 |
4054 | ### Response Timing
4055 |
4056 | * The first response might take up to 30 seconds to return
4057 | * This is normal and happens while:
4058 | * The server initializes
4059 | * Claude processes the query
4060 | * Tools are being executed
4061 | * Subsequent responses are typically faster
4062 | * Don't interrupt the process during this initial waiting period
4063 |
4064 | ### Common Error Messages
4065 |
4066 | If you see:
4067 |
4068 | * `Error: Cannot find module`: Check your build folder and ensure TypeScript compilation succeeded
4069 | * `Connection refused`: Ensure the server is running and the path is correct
4070 | * `Tool execution failed`: Verify the tool's required environment variables are set
4071 | * `ANTHROPIC_API_KEY is not set`: Check your .env file and environment variables
4072 | * `TypeError`: Ensure you're using the correct types for tool arguments
4073 | </Tab>
4074 |
4075 | <Tab title="Java">
4076 | <Note>
4077 | This is a quickstart demo based on Spring AI MCP auto-configuration and boot starters.
4078 | To learn how to create sync and async MCP Clients manually, consult the [Java SDK Client](/sdk/java/mcp-client) documentation
4079 | </Note>
4080 |
4081 | This example demonstrates how to build an interactive chatbot that combines Spring AI's Model Context Protocol (MCP) with the [Brave Search MCP Server](https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search). The application creates a conversational interface powered by Anthropic's Claude AI model that can perform internet searches through Brave Search, enabling natural language interactions with real-time web data.
4082 | [You can find the complete code for this tutorial here.](https://github.com/spring-projects/spring-ai-examples/tree/main/model-context-protocol/web-search/brave-chatbot)
4083 |
4084 | ## System Requirements
4085 |
4086 | Before starting, ensure your system meets these requirements:
4087 |
4088 | * Java 17 or higher
4089 | * Maven 3.6+
4090 | * npx package manager
4091 | * Anthropic API key (Claude)
4092 | * Brave Search API key
4093 |
4094 | ## Setting Up Your Environment
4095 |
4096 | 1. Install npx (Node Package eXecute):
4097 | First, make sure to install [npm](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm)
4098 | and then run:
4099 | ```bash
4100 | npm install -g npx
4101 | ```
4102 |
4103 | 2. Clone the repository:
4104 | ```bash
4105 | git clone https://github.com/spring-projects/spring-ai-examples.git
4106 | cd model-context-protocol/brave-chatbot
4107 | ```
4108 |
4109 | 3. Set up your API keys:
4110 | ```bash
4111 | export ANTHROPIC_API_KEY='your-anthropic-api-key-here'
4112 | export BRAVE_API_KEY='your-brave-api-key-here'
4113 | ```
4114 |
4115 | 4. Build the application:
4116 | ```bash
4117 | ./mvnw clean install
4118 | ```
4119 |
4120 | 5. Run the application using Maven:
4121 | ```bash
4122 | ./mvnw spring-boot:run
4123 | ```
4124 |
4125 | <Warning>
4126 | Make sure you keep your `ANTHROPIC_API_KEY` and `BRAVE_API_KEY` keys secure!
4127 | </Warning>
4128 |
4129 | ## How it Works
4130 |
4131 | The application integrates Spring AI with the Brave Search MCP server through several components:
4132 |
4133 | ### MCP Client Configuration
4134 |
4135 | 1. Required dependencies in pom.xml:
4136 |
4137 | ```xml
4138 | <dependency>
4139 | <groupId>org.springframework.ai</groupId>
4140 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
4141 | </dependency>
4142 | <dependency>
4143 | <groupId>org.springframework.ai</groupId>
4144 | <artifactId>spring-ai-anthropic-spring-boot-starter</artifactId>
4145 | </dependency>
4146 | ```
4147 |
4148 | 2. Application properties (application.yml):
4149 |
4150 | ```yml
4151 | spring:
4152 | ai:
4153 | mcp:
4154 | client:
4155 | enabled: true
4156 | name: brave-search-client
4157 | version: 1.0.0
4158 | type: SYNC
4159 | request-timeout: 20s
4160 | stdio:
4161 | root-change-notification: true
4162 | servers-configuration: classpath:/mcp-servers-config.json
4163 | anthropic:
4164 | api-key: ${ANTHROPIC_API_KEY}
4165 | ```
4166 |
4167 | This activates the `spring-ai-mcp-client-spring-boot-starter` to create one or more `McpClient`s based on the provided server configuration.
4168 |
4169 | 3. MCP Server Configuration (`mcp-servers-config.json`):
4170 |
4171 | ```json
4172 | {
4173 | "mcpServers": {
4174 | "brave-search": {
4175 | "command": "npx",
4176 | "args": [
4177 | "-y",
4178 | "@modelcontextprotocol/server-brave-search"
4179 | ],
4180 | "env": {
4181 | "BRAVE_API_KEY": "<PUT YOUR BRAVE API KEY>"
4182 | }
4183 | }
4184 | }
4185 | }
4186 | ```
4187 |
4188 | ### Chat Implementation
4189 |
4190 | The chatbot is implemented using Spring AI's ChatClient with MCP tool integration:
4191 |
4192 | ```java
4193 | var chatClient = chatClientBuilder
4194 | .defaultSystem("You are useful assistant, expert in AI and Java.")
4195 | .defaultTools((Object[]) mcpToolAdapter.toolCallbacks())
4196 | .defaultAdvisors(new MessageChatMemoryAdvisor(new InMemoryChatMemory()))
4197 | .build();
4198 | ```
4199 |
4200 | Key features:
4201 |
4202 | * Uses Claude AI model for natural language understanding
4203 | * Integrates Brave Search through MCP for real-time web search capabilities
4204 | * Maintains conversation memory using InMemoryChatMemory
4205 | * Runs as an interactive command-line application
4206 |
4207 | ### Build and run
4208 |
4209 | ```bash
4210 | ./mvnw clean install
4211 | java -jar ./target/ai-mcp-brave-chatbot-0.0.1-SNAPSHOT.jar
4212 | ```
4213 |
4214 | or
4215 |
4216 | ```bash
4217 | ./mvnw spring-boot:run
4218 | ```
4219 |
4220 | The application will start an interactive chat session where you can ask questions. The chatbot will use Brave Search when it needs to find information from the internet to answer your queries.
4221 |
4222 | The chatbot can:
4223 |
4224 | * Answer questions using its built-in knowledge
4225 | * Perform web searches when needed using Brave Search
4226 | * Remember context from previous messages in the conversation
4227 | * Combine information from multiple sources to provide comprehensive answers
4228 |
4229 | ### Advanced Configuration
4230 |
4231 | The MCP client supports additional configuration options:
4232 |
4233 | * Client customization through `McpSyncClientCustomizer` or `McpAsyncClientCustomizer`
4234 | * Multiple clients with multiple transport types: `STDIO` and `SSE` (Server-Sent Events)
4235 | * Integration with Spring AI's tool execution framework
4236 | * Automatic client initialization and lifecycle management
4237 |
4238 | For WebFlux-based applications, you can use the WebFlux starter instead:
4239 |
4240 | ```xml
4241 | <dependency>
4242 | <groupId>org.springframework.ai</groupId>
4243 | <artifactId>spring-ai-mcp-client-webflux-spring-boot-starter</artifactId>
4244 | </dependency>
4245 | ```
4246 |
4247 | This provides similar functionality but uses a WebFlux-based SSE transport implementation, recommended for production deployments.
4248 | </Tab>
4249 | </Tabs>
4250 |
4251 | ## Next steps
4252 |
4253 | <CardGroup cols={2}>
4254 | <Card title="Example servers" icon="grid" href="/examples">
4255 | Check out our gallery of official MCP servers and implementations
4256 | </Card>
4257 |
4258 | <Card title="Clients" icon="cubes" href="/clients">
4259 | View the list of clients that support MCP integrations
4260 | </Card>
4261 |
4262 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
4263 | Learn how to use LLMs like Claude to speed up your MCP development
4264 | </Card>
4265 |
4266 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
4267 | Understand how MCP connects clients, servers, and LLMs
4268 | </Card>
4269 | </CardGroup>
4270 |
4271 |
4272 | # For Server Developers
4273 | Source: https://modelcontextprotocol.io/quickstart/server
4274 |
4275 | Get started building your own server to use in Claude for Desktop and other clients.
4276 |
4277 | In this tutorial, we'll build a simple MCP weather server and connect it to a host, Claude for Desktop. We'll start with a basic setup, and then progress to more complex use cases.
4278 |
4279 | ### What we'll be building
4280 |
4281 | Many LLMs (including Claude) do not currently have the ability to fetch the forecast and severe weather alerts. Let's use MCP to solve that!
4282 |
4283 | We'll build a server that exposes two tools: `get-alerts` and `get-forecast`. Then we'll connect the server to an MCP host (in this case, Claude for Desktop):
4284 |
4285 | <Frame>
4286 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
4287 | </Frame>
4288 |
4289 | <Frame>
4290 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
4291 | </Frame>
4292 |
4293 | <Note>
4294 | Servers can connect to any client. We've chosen Claude for Desktop here for simplicity, but we also have guides on [building your own client](/quickstart/client) as well as a [list of other clients here](/clients).
4295 | </Note>
4296 |
4297 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
4298 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
4299 | </Accordion>
4300 |
4301 | ### Core MCP Concepts
4302 |
4303 | MCP servers can provide three main types of capabilities:
4304 |
4305 | 1. **Resources**: File-like data that can be read by clients (like API responses or file contents)
4306 | 2. **Tools**: Functions that can be called by the LLM (with user approval)
4307 | 3. **Prompts**: Pre-written templates that help users accomplish specific tasks
4308 |
4309 | This tutorial will primarily focus on tools.
4310 |
4311 | <Tabs>
4312 | <Tab title="Python">
4313 | Let's get started with building our weather server! [You can find the complete code for what we'll be building here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/weather-server-python)
4314 |
4315 | ### Prerequisite knowledge
4316 |
4317 | This quickstart assumes you have familiarity with:
4318 |
4319 | * Python
4320 | * LLMs like Claude
4321 |
4322 | ### System requirements
4323 |
4324 | * Python 3.10 or higher installed.
4325 | * You must use the Python MCP SDK 1.2.0 or higher.
4326 |
4327 | ### Set up your environment
4328 |
4329 | First, let's install `uv` and set up our Python project and environment:
4330 |
4331 | <CodeGroup>
4332 | ```bash MacOS/Linux
4333 | curl -LsSf https://astral.sh/uv/install.sh | sh
4334 | ```
4335 |
4336 | ```powershell Windows
4337 | powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
4338 | ```
4339 | </CodeGroup>
4340 |
4341 | Make sure to restart your terminal afterwards to ensure that the `uv` command gets picked up.
4342 |
4343 | Now, let's create and set up our project:
4344 |
4345 | <CodeGroup>
4346 | ```bash MacOS/Linux
4347 | # Create a new directory for our project
4348 | uv init weather
4349 | cd weather
4350 |
4351 | # Create virtual environment and activate it
4352 | uv venv
4353 | source .venv/bin/activate
4354 |
4355 | # Install dependencies
4356 | uv add "mcp[cli]" httpx
4357 |
4358 | # Create our server file
4359 | touch weather.py
4360 | ```
4361 |
4362 | ```powershell Windows
4363 | # Create a new directory for our project
4364 | uv init weather
4365 | cd weather
4366 |
4367 | # Create virtual environment and activate it
4368 | uv venv
4369 | .venv\Scripts\activate
4370 |
4371 | # Install dependencies
4372 | uv add mcp[cli] httpx
4373 |
4374 | # Create our server file
4375 | new-item weather.py
4376 | ```
4377 | </CodeGroup>
4378 |
4379 | Now let's dive into building your server.
4380 |
4381 | ## Building your server
4382 |
4383 | ### Importing packages and setting up the instance
4384 |
4385 | Add these to the top of your `weather.py`:
4386 |
4387 | ```python
4388 | from typing import Any
4389 | import httpx
4390 | from mcp.server.fastmcp import FastMCP
4391 |
4392 | # Initialize FastMCP server
4393 | mcp = FastMCP("weather")
4394 |
4395 | # Constants
4396 | NWS_API_BASE = "https://api.weather.gov"
4397 | USER_AGENT = "weather-app/1.0"
4398 | ```
4399 |
4400 | The FastMCP class uses Python type hints and docstrings to automatically generate tool definitions, making it easy to create and maintain MCP tools.
4401 |
4402 | ### Helper functions
4403 |
4404 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
4405 |
4406 | ```python
4407 | async def make_nws_request(url: str) -> dict[str, Any] | None:
4408 | """Make a request to the NWS API with proper error handling."""
4409 | headers = {
4410 | "User-Agent": USER_AGENT,
4411 | "Accept": "application/geo+json"
4412 | }
4413 | async with httpx.AsyncClient() as client:
4414 | try:
4415 | response = await client.get(url, headers=headers, timeout=30.0)
4416 | response.raise_for_status()
4417 | return response.json()
4418 | except Exception:
4419 | return None
4420 |
4421 | def format_alert(feature: dict) -> str:
4422 | """Format an alert feature into a readable string."""
4423 | props = feature["properties"]
4424 | return f"""
4425 | Event: {props.get('event', 'Unknown')}
4426 | Area: {props.get('areaDesc', 'Unknown')}
4427 | Severity: {props.get('severity', 'Unknown')}
4428 | Description: {props.get('description', 'No description available')}
4429 | Instructions: {props.get('instruction', 'No specific instructions provided')}
4430 | """
4431 | ```
4432 |
4433 | ### Implementing tool execution
4434 |
4435 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4436 |
4437 | ```python
4438 | @mcp.tool()
4439 | async def get_alerts(state: str) -> str:
4440 | """Get weather alerts for a US state.
4441 |
4442 | Args:
4443 | state: Two-letter US state code (e.g. CA, NY)
4444 | """
4445 | url = f"{NWS_API_BASE}/alerts/active/area/{state}"
4446 | data = await make_nws_request(url)
4447 |
4448 | if not data or "features" not in data:
4449 | return "Unable to fetch alerts or no alerts found."
4450 |
4451 | if not data["features"]:
4452 | return "No active alerts for this state."
4453 |
4454 | alerts = [format_alert(feature) for feature in data["features"]]
4455 | return "\n---\n".join(alerts)
4456 |
4457 | @mcp.tool()
4458 | async def get_forecast(latitude: float, longitude: float) -> str:
4459 | """Get weather forecast for a location.
4460 |
4461 | Args:
4462 | latitude: Latitude of the location
4463 | longitude: Longitude of the location
4464 | """
4465 | # First get the forecast grid endpoint
4466 | points_url = f"{NWS_API_BASE}/points/{latitude},{longitude}"
4467 | points_data = await make_nws_request(points_url)
4468 |
4469 | if not points_data:
4470 | return "Unable to fetch forecast data for this location."
4471 |
4472 | # Get the forecast URL from the points response
4473 | forecast_url = points_data["properties"]["forecast"]
4474 | forecast_data = await make_nws_request(forecast_url)
4475 |
4476 | if not forecast_data:
4477 | return "Unable to fetch detailed forecast."
4478 |
4479 | # Format the periods into a readable forecast
4480 | periods = forecast_data["properties"]["periods"]
4481 | forecasts = []
4482 | for period in periods[:5]: # Only show next 5 periods
4483 | forecast = f"""
4484 | {period['name']}:
4485 | Temperature: {period['temperature']}°{period['temperatureUnit']}
4486 | Wind: {period['windSpeed']} {period['windDirection']}
4487 | Forecast: {period['detailedForecast']}
4488 | """
4489 | forecasts.append(forecast)
4490 |
4491 | return "\n---\n".join(forecasts)
4492 | ```
4493 |
4494 | ### Running the server
4495 |
4496 | Finally, let's initialize and run the server:
4497 |
4498 | ```python
4499 | if __name__ == "__main__":
4500 | # Initialize and run the server
4501 | mcp.run(transport='stdio')
4502 | ```
4503 |
4504 | Your server is complete! Run `uv run weather.py` to confirm that everything's working.
4505 |
4506 | Let's now test your server from an existing MCP host, Claude for Desktop.
4507 |
4508 | ## Testing your server with Claude for Desktop
4509 |
4510 | <Note>
4511 | Claude for Desktop is not yet available on Linux. Linux users can proceed to the [Building a client](/quickstart/client) tutorial to build an MCP client that connects to the server we just built.
4512 | </Note>
4513 |
4514 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4515 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4516 |
4517 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use. To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor. Make sure to create the file if it doesn't exist.
4518 |
4519 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4520 |
4521 | <Tabs>
4522 | <Tab title="MacOS/Linux">
4523 | ```bash
4524 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4525 | ```
4526 | </Tab>
4527 |
4528 | <Tab title="Windows">
4529 | ```powershell
4530 | code $env:AppData\Claude\claude_desktop_config.json
4531 | ```
4532 | </Tab>
4533 | </Tabs>
4534 |
4535 | You'll then add your servers in the `mcpServers` key. The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4536 |
4537 | In this case, we'll add our single weather server like so:
4538 |
4539 | <Tabs>
4540 | <Tab title="MacOS/Linux">
4541 | ```json Python
4542 | {
4543 | "mcpServers": {
4544 | "weather": {
4545 | "command": "uv",
4546 | "args": [
4547 | "--directory",
4548 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather",
4549 | "run",
4550 | "weather.py"
4551 | ]
4552 | }
4553 | }
4554 | }
4555 | ```
4556 | </Tab>
4557 |
4558 | <Tab title="Windows">
4559 | ```json Python
4560 | {
4561 | "mcpServers": {
4562 | "weather": {
4563 | "command": "uv",
4564 | "args": [
4565 | "--directory",
4566 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather",
4567 | "run",
4568 | "weather.py"
4569 | ]
4570 | }
4571 | }
4572 | }
4573 | ```
4574 | </Tab>
4575 | </Tabs>
4576 |
4577 | <Warning>
4578 | You may need to put the full path to the `uv` executable in the `command` field. You can get this by running `which uv` on MacOS/Linux or `where uv` on Windows.
4579 | </Warning>
4580 |
4581 | <Note>
4582 | Make sure you pass in the absolute path to your server.
4583 | </Note>
4584 |
4585 | This tells Claude for Desktop:
4586 |
4587 | 1. There's an MCP server named "weather"
4588 | 2. To launch it by running `uv --directory /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather run weather.py`
4589 |
4590 | Save the file, and restart **Claude for Desktop**.
4591 | </Tab>
4592 |
4593 | <Tab title="Node">
4594 | Let's get started with building our weather server! [You can find the complete code for what we'll be building here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/weather-server-typescript)
4595 |
4596 | ### Prerequisite knowledge
4597 |
4598 | This quickstart assumes you have familiarity with:
4599 |
4600 | * TypeScript
4601 | * LLMs like Claude
4602 |
4603 | ### System requirements
4604 |
4605 | For TypeScript, make sure you have the latest version of Node installed.
4606 |
4607 | ### Set up your environment
4608 |
4609 | First, let's install Node.js and npm if you haven't already. You can download them from [nodejs.org](https://nodejs.org/).
4610 | Verify your Node.js installation:
4611 |
4612 | ```bash
4613 | node --version
4614 | npm --version
4615 | ```
4616 |
4617 | For this tutorial, you'll need Node.js version 16 or higher.
4618 |
4619 | Now, let's create and set up our project:
4620 |
4621 | <CodeGroup>
4622 | ```bash MacOS/Linux
4623 | # Create a new directory for our project
4624 | mkdir weather
4625 | cd weather
4626 |
4627 | # Initialize a new npm project
4628 | npm init -y
4629 |
4630 | # Install dependencies
4631 | npm install @modelcontextprotocol/sdk zod
4632 | npm install -D @types/node typescript
4633 |
4634 | # Create our files
4635 | mkdir src
4636 | touch src/index.ts
4637 | ```
4638 |
4639 | ```powershell Windows
4640 | # Create a new directory for our project
4641 | md weather
4642 | cd weather
4643 |
4644 | # Initialize a new npm project
4645 | npm init -y
4646 |
4647 | # Install dependencies
4648 | npm install @modelcontextprotocol/sdk zod
4649 | npm install -D @types/node typescript
4650 |
4651 | # Create our files
4652 | md src
4653 | new-item src\index.ts
4654 | ```
4655 | </CodeGroup>
4656 |
4657 | Update your package.json to add type: "module" and a build script:
4658 |
4659 | ```json package.json
4660 | {
4661 | "type": "module",
4662 | "bin": {
4663 | "weather": "./build/index.js"
4664 | },
4665 | "scripts": {
4666 | "build": "tsc && chmod 755 build/index.js"
4667 | },
4668 | "files": [
4669 | "build"
4670 | ],
4671 | }
4672 | ```
4673 |
4674 | Create a `tsconfig.json` in the root of your project:
4675 |
4676 | ```json tsconfig.json
4677 | {
4678 | "compilerOptions": {
4679 | "target": "ES2022",
4680 | "module": "Node16",
4681 | "moduleResolution": "Node16",
4682 | "outDir": "./build",
4683 | "rootDir": "./src",
4684 | "strict": true,
4685 | "esModuleInterop": true,
4686 | "skipLibCheck": true,
4687 | "forceConsistentCasingInFileNames": true
4688 | },
4689 | "include": ["src/**/*"],
4690 | "exclude": ["node_modules"]
4691 | }
4692 | ```
4693 |
4694 | Now let's dive into building your server.
4695 |
4696 | ## Building your server
4697 |
4698 | ### Importing packages and setting up the instance
4699 |
4700 | Add these to the top of your `src/index.ts`:
4701 |
4702 | ```typescript
4703 | import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
4704 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
4705 | import { z } from "zod";
4706 |
4707 | const NWS_API_BASE = "https://api.weather.gov";
4708 | const USER_AGENT = "weather-app/1.0";
4709 |
4710 | // Create server instance
4711 | const server = new McpServer({
4712 | name: "weather",
4713 | version: "1.0.0",
4714 | });
4715 | ```
4716 |
4717 | ### Helper functions
4718 |
4719 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
4720 |
4721 | ```typescript
4722 | // Helper function for making NWS API requests
4723 | async function makeNWSRequest<T>(url: string): Promise<T | null> {
4724 | const headers = {
4725 | "User-Agent": USER_AGENT,
4726 | Accept: "application/geo+json",
4727 | };
4728 |
4729 | try {
4730 | const response = await fetch(url, { headers });
4731 | if (!response.ok) {
4732 | throw new Error(`HTTP error! status: ${response.status}`);
4733 | }
4734 | return (await response.json()) as T;
4735 | } catch (error) {
4736 | console.error("Error making NWS request:", error);
4737 | return null;
4738 | }
4739 | }
4740 |
4741 | interface AlertFeature {
4742 | properties: {
4743 | event?: string;
4744 | areaDesc?: string;
4745 | severity?: string;
4746 | status?: string;
4747 | headline?: string;
4748 | };
4749 | }
4750 |
4751 | // Format alert data
4752 | function formatAlert(feature: AlertFeature): string {
4753 | const props = feature.properties;
4754 | return [
4755 | `Event: ${props.event || "Unknown"}`,
4756 | `Area: ${props.areaDesc || "Unknown"}`,
4757 | `Severity: ${props.severity || "Unknown"}`,
4758 | `Status: ${props.status || "Unknown"}`,
4759 | `Headline: ${props.headline || "No headline"}`,
4760 | "---",
4761 | ].join("\n");
4762 | }
4763 |
4764 | interface ForecastPeriod {
4765 | name?: string;
4766 | temperature?: number;
4767 | temperatureUnit?: string;
4768 | windSpeed?: string;
4769 | windDirection?: string;
4770 | shortForecast?: string;
4771 | }
4772 |
4773 | interface AlertsResponse {
4774 | features: AlertFeature[];
4775 | }
4776 |
4777 | interface PointsResponse {
4778 | properties: {
4779 | forecast?: string;
4780 | };
4781 | }
4782 |
4783 | interface ForecastResponse {
4784 | properties: {
4785 | periods: ForecastPeriod[];
4786 | };
4787 | }
4788 | ```
4789 |
4790 | ### Implementing tool execution
4791 |
4792 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4793 |
4794 | ```typescript
4795 | // Register weather tools
4796 | server.tool(
4797 | "get-alerts",
4798 | "Get weather alerts for a state",
4799 | {
4800 | state: z.string().length(2).describe("Two-letter state code (e.g. CA, NY)"),
4801 | },
4802 | async ({ state }) => {
4803 | const stateCode = state.toUpperCase();
4804 | const alertsUrl = `${NWS_API_BASE}/alerts?area=${stateCode}`;
4805 | const alertsData = await makeNWSRequest<AlertsResponse>(alertsUrl);
4806 |
4807 | if (!alertsData) {
4808 | return {
4809 | content: [
4810 | {
4811 | type: "text",
4812 | text: "Failed to retrieve alerts data",
4813 | },
4814 | ],
4815 | };
4816 | }
4817 |
4818 | const features = alertsData.features || [];
4819 | if (features.length === 0) {
4820 | return {
4821 | content: [
4822 | {
4823 | type: "text",
4824 | text: `No active alerts for ${stateCode}`,
4825 | },
4826 | ],
4827 | };
4828 | }
4829 |
4830 | const formattedAlerts = features.map(formatAlert);
4831 | const alertsText = `Active alerts for ${stateCode}:\n\n${formattedAlerts.join("\n")}`;
4832 |
4833 | return {
4834 | content: [
4835 | {
4836 | type: "text",
4837 | text: alertsText,
4838 | },
4839 | ],
4840 | };
4841 | },
4842 | );
4843 |
4844 | server.tool(
4845 | "get-forecast",
4846 | "Get weather forecast for a location",
4847 | {
4848 | latitude: z.number().min(-90).max(90).describe("Latitude of the location"),
4849 | longitude: z.number().min(-180).max(180).describe("Longitude of the location"),
4850 | },
4851 | async ({ latitude, longitude }) => {
4852 | // Get grid point data
4853 | const pointsUrl = `${NWS_API_BASE}/points/${latitude.toFixed(4)},${longitude.toFixed(4)}`;
4854 | const pointsData = await makeNWSRequest<PointsResponse>(pointsUrl);
4855 |
4856 | if (!pointsData) {
4857 | return {
4858 | content: [
4859 | {
4860 | type: "text",
4861 | text: `Failed to retrieve grid point data for coordinates: ${latitude}, ${longitude}. This location may not be supported by the NWS API (only US locations are supported).`,
4862 | },
4863 | ],
4864 | };
4865 | }
4866 |
4867 | const forecastUrl = pointsData.properties?.forecast;
4868 | if (!forecastUrl) {
4869 | return {
4870 | content: [
4871 | {
4872 | type: "text",
4873 | text: "Failed to get forecast URL from grid point data",
4874 | },
4875 | ],
4876 | };
4877 | }
4878 |
4879 | // Get forecast data
4880 | const forecastData = await makeNWSRequest<ForecastResponse>(forecastUrl);
4881 | if (!forecastData) {
4882 | return {
4883 | content: [
4884 | {
4885 | type: "text",
4886 | text: "Failed to retrieve forecast data",
4887 | },
4888 | ],
4889 | };
4890 | }
4891 |
4892 | const periods = forecastData.properties?.periods || [];
4893 | if (periods.length === 0) {
4894 | return {
4895 | content: [
4896 | {
4897 | type: "text",
4898 | text: "No forecast periods available",
4899 | },
4900 | ],
4901 | };
4902 | }
4903 |
4904 | // Format forecast periods
4905 | const formattedForecast = periods.map((period: ForecastPeriod) =>
4906 | [
4907 | `${period.name || "Unknown"}:`,
4908 | `Temperature: ${period.temperature || "Unknown"}°${period.temperatureUnit || "F"}`,
4909 | `Wind: ${period.windSpeed || "Unknown"} ${period.windDirection || ""}`,
4910 | `${period.shortForecast || "No forecast available"}`,
4911 | "---",
4912 | ].join("\n"),
4913 | );
4914 |
4915 | const forecastText = `Forecast for ${latitude}, ${longitude}:\n\n${formattedForecast.join("\n")}`;
4916 |
4917 | return {
4918 | content: [
4919 | {
4920 | type: "text",
4921 | text: forecastText,
4922 | },
4923 | ],
4924 | };
4925 | },
4926 | );
4927 | ```
4928 |
4929 | ### Running the server
4930 |
4931 | Finally, implement the main function to run the server:
4932 |
4933 | ```typescript
4934 | async function main() {
4935 | const transport = new StdioServerTransport();
4936 | await server.connect(transport);
4937 | console.error("Weather MCP Server running on stdio");
4938 | }
4939 |
4940 | main().catch((error) => {
4941 | console.error("Fatal error in main():", error);
4942 | process.exit(1);
4943 | });
4944 | ```
4945 |
4946 | Make sure to run `npm run build` to build your server! This is a very important step in getting your server to connect.
4947 |
4948 | Let's now test your server from an existing MCP host, Claude for Desktop.
4949 |
4950 | ## Testing your server with Claude for Desktop
4951 |
4952 | <Note>
4953 | Claude for Desktop is not yet available on Linux. Linux users can proceed to the [Building a client](/quickstart/client) tutorial to build an MCP client that connects to the server we just built.
4954 | </Note>
4955 |
4956 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4957 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4958 |
4959 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use. To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor. Make sure to create the file if it doesn't exist.
4960 |
4961 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4962 |
4963 | <Tabs>
4964 | <Tab title="MacOS/Linux">
4965 | ```bash
4966 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4967 | ```
4968 | </Tab>
4969 |
4970 | <Tab title="Windows">
4971 | ```powershell
4972 | code $env:AppData\Claude\claude_desktop_config.json
4973 | ```
4974 | </Tab>
4975 | </Tabs>
4976 |
4977 | You'll then add your servers in the `mcpServers` key. The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4978 |
4979 | In this case, we'll add our single weather server like so:
4980 |
4981 | <Tabs>
4982 | <Tab title="MacOS/Linux">
4983 | <CodeGroup>
4984 | ```json Node
4985 | {
4986 | "mcpServers": {
4987 | "weather": {
4988 | "command": "node",
4989 | "args": [
4990 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js"
4991 | ]
4992 | }
4993 | }
4994 | }
4995 | ```
4996 | </CodeGroup>
4997 | </Tab>
4998 |
4999 | <Tab title="Windows">
5000 | <CodeGroup>
5001 | ```json Node
5002 | {
5003 | "mcpServers": {
5004 | "weather": {
5005 | "command": "node",
5006 | "args": [
5007 | "C:\\PATH\\TO\\PARENT\\FOLDER\\weather\\build\\index.js"
5008 | ]
5009 | }
5010 | }
5011 | }
5012 | ```
5013 | </CodeGroup>
5014 | </Tab>
5015 | </Tabs>
5016 |
5017 | This tells Claude for Desktop:
5018 |
5019 | 1. There's an MCP server named "weather"
5020 | 2. Launch it by running `node /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js`
5021 |
5022 | Save the file, and restart **Claude for Desktop**.
5023 | </Tab>
5024 |
5025 | <Tab title="Java">
5026 | <Note>
5027 | This is a quickstart demo based on Spring AI MCP auto-configuration and boot starters.
5028 | To learn how to create sync and async MCP Servers, manually, consult the [Java SDK Server](/sdk/java/mcp-server) documentation.
5029 | </Note>
5030 |
5031 | Let's get started with building our weather server!
5032 | [You can find the complete code for what we'll be building here.](https://github.com/spring-projects/spring-ai-examples/tree/main/model-context-protocol/weather/starter-stdio-server)
5033 |
5034 | For more information, see the [MCP Server Boot Starter](https://docs.spring.io/spring-ai/reference/api/mcp/mcp-server-boot-starter-docs.html) reference documentation.
5035 | For manual MCP Server implementation, refer to the [MCP Server Java SDK documentation](/sdk/java/mcp-server).
5036 |
5037 | ### System requirements
5038 |
5039 | * Java 17 or higher installed.
5040 | * [Spring Boot 3.3.x](https://docs.spring.io/spring-boot/installing.html) or higher
5041 |
5042 | ### Set up your environment
5043 |
5044 | Use the [Spring Initizer](https://start.spring.io/) to bootstrat the project.
5045 |
5046 | You will need to add the following dependencies:
5047 |
5048 | <Tabs>
5049 | <Tab title="Maven">
5050 | ```xml
5051 | <dependencies>
5052 | <dependency>
5053 | <groupId>org.springframework.ai</groupId>
5054 | <artifactId>spring-ai-mcp-server-spring-boot-starter</artifactId>
5055 | </dependency>
5056 |
5057 | <dependency>
5058 | <groupId>org.springframework</groupId>
5059 | <artifactId>spring-web</artifactId>
5060 | </dependency>
5061 | </dependencies>
5062 | ```
5063 | </Tab>
5064 |
5065 | <Tab title="Gradle">
5066 | ```groovy
5067 | dependencies {
5068 | implementation platform("org.springframework.ai:spring-ai-mcp-server-spring-boot-starter")
5069 | implementation platform("org.springframework:spring-web")
5070 | }
5071 | ```
5072 | </Tab>
5073 | </Tabs>
5074 |
5075 | Then configure your application by setting the applicaiton properties:
5076 |
5077 | <CodeGroup>
5078 | ```bash application.properties
5079 | spring.main.bannerMode=off
5080 | logging.pattern.console=
5081 | ```
5082 |
5083 | ```yaml application.yml
5084 | logging:
5085 | pattern:
5086 | console:
5087 | spring:
5088 | main:
5089 | banner-mode: off
5090 | ```
5091 | </CodeGroup>
5092 |
5093 | The [Server Configuration Properties](https://docs.spring.io/spring-ai/reference/api/mcp/mcp-server-boot-starter-docs.html#_configuration_properties) documents all available properties.
5094 |
5095 | Now let's dive into building your server.
5096 |
5097 | ## Building your server
5098 |
5099 | ### Weather Service
5100 |
5101 | Let's implement a [WeatheService.java](https://github.com/spring-projects/spring-ai-examples/blob/main/model-context-protocol/weather/starter-stdio-server/src/main/java/org/springframework/ai/mcp/sample/server/WeatherService.java) that uses a REST client to query the data from the National Weather Service API:
5102 |
5103 | ```java
5104 | @Service
5105 | public class WeatherService {
5106 |
5107 | private final RestClient restClient;
5108 |
5109 | public WeatherService() {
5110 | this.restClient = RestClient.builder()
5111 | .baseUrl("https://api.weather.gov")
5112 | .defaultHeader("Accept", "application/geo+json")
5113 | .defaultHeader("User-Agent", "WeatherApiClient/1.0 ([email protected])")
5114 | .build();
5115 | }
5116 |
5117 | @Tool(description = "Get weather forecast for a specific latitude/longitude")
5118 | public String getWeatherForecastByLocation(
5119 | double latitude, // Latitude coordinate
5120 | double longitude // Longitude coordinate
5121 | ) {
5122 | // Returns detailed forecast including:
5123 | // - Temperature and unit
5124 | // - Wind speed and direction
5125 | // - Detailed forecast description
5126 | }
5127 |
5128 | @Tool(description = "Get weather alerts for a US state")
5129 | public String getAlerts(
5130 | @ToolParam(description = "Two-letter US state code (e.g. CA, NY") String state)
5131 | ) {
5132 | // Returns active alerts including:
5133 | // - Event type
5134 | // - Affected area
5135 | // - Severity
5136 | // - Description
5137 | // - Safety instructions
5138 | }
5139 |
5140 | // ......
5141 | }
5142 | ```
5143 |
5144 | The `@Service` annotation with auto-register the service in your applicaiton context.
5145 | The Spring AI `@Tool` annotation, making it easy to create and maintain MCP tools.
5146 |
5147 | The auto-configuration will automatically register these tools with the MCP server.
5148 |
5149 | ### Create your Boot Applicaiton
5150 |
5151 | ```java
5152 | @SpringBootApplication
5153 | public class McpServerApplication {
5154 |
5155 | public static void main(String[] args) {
5156 | SpringApplication.run(McpServerApplication.class, args);
5157 | }
5158 |
5159 | @Bean
5160 | public ToolCallbackProvider weatherTools(WeatherService weatherService) {
5161 | return MethodToolCallbackProvider.builder().toolObjects(weatherService).build();
5162 | }
5163 | }
5164 | ```
5165 |
5166 | Uses the the `MethodToolCallbackProvider` utils to convert the `@Tools` into actionalble callbackes used by the MCP server.
5167 |
5168 | ### Running the server
5169 |
5170 | Finally, let's build the server:
5171 |
5172 | ```bash
5173 | ./mvnw clean install
5174 | ```
5175 |
5176 | This will generate a `mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar` file within the `target` folder.
5177 |
5178 | Let's now test your server from an existing MCP host, Claude for Desktop.
5179 |
5180 | ## Testing your server with Claude for Desktop
5181 |
5182 | <Note>
5183 | Claude for Desktop is not yet available on Linux.
5184 | </Note>
5185 |
5186 | First, make sure you have Claude for Desktop installed.
5187 | [You can install the latest version here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
5188 |
5189 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use.
5190 | To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor.
5191 | Make sure to create the file if it doesn't exist.
5192 |
5193 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
5194 |
5195 | <Tabs>
5196 | <Tab title="MacOS/Linux">
5197 | ```bash
5198 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
5199 | ```
5200 | </Tab>
5201 |
5202 | <Tab title="Windows">
5203 | ```powershell
5204 | code $env:AppData\Claude\claude_desktop_config.json
5205 | ```
5206 | </Tab>
5207 | </Tabs>
5208 |
5209 | You'll then add your servers in the `mcpServers` key.
5210 | The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
5211 |
5212 | In this case, we'll add our single weather server like so:
5213 |
5214 | <Tabs>
5215 | <Tab title="MacOS/Linux">
5216 | ```json java
5217 | {
5218 | "mcpServers": {
5219 | "spring-ai-mcp-weather": {
5220 | "command": "java",
5221 | "args": [
5222 | "-Dspring.ai.mcp.server.stdio=true",
5223 | "-jar",
5224 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
5225 | ]
5226 | }
5227 | }
5228 | }
5229 | ```
5230 | </Tab>
5231 |
5232 | <Tab title="Windows">
5233 | ```json java
5234 | {
5235 | "mcpServers": {
5236 | "spring-ai-mcp-weather": {
5237 | "command": "java",
5238 | "args": [
5239 | "-Dspring.ai.mcp.server.transport=STDIO",
5240 | "-jar",
5241 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather\\mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
5242 | ]
5243 | }
5244 | }
5245 | }
5246 | ```
5247 | </Tab>
5248 | </Tabs>
5249 |
5250 | <Note>
5251 | Make sure you pass in the absolute path to your server.
5252 | </Note>
5253 |
5254 | This tells Claude for Desktop:
5255 |
5256 | 1. There's an MCP server named "my-weather-server"
5257 | 2. To launch it by running `java -jar /ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar`
5258 |
5259 | Save the file, and restart **Claude for Desktop**.
5260 |
5261 | ## Testing your server with Java client
5262 |
5263 | ### Create a MCP Client manually
5264 |
5265 | Use the `McpClient` to connect to the server:
5266 |
5267 | ```java
5268 | var stdioParams = ServerParameters.builder("java")
5269 | .args("-jar", "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar")
5270 | .build();
5271 |
5272 | var stdioTransport = new StdioClientTransport(stdioParams);
5273 |
5274 | var mcpClient = McpClient.sync(stdioTransport).build();
5275 |
5276 | mcpClient.initialize();
5277 |
5278 | ListToolsResult toolsList = mcpClient.listTools();
5279 |
5280 | CallToolResult weather = mcpClient.callTool(
5281 | new CallToolRequest("getWeatherForecastByLocation",
5282 | Map.of("latitude", "47.6062", "longitude", "-122.3321")));
5283 |
5284 | CallToolResult alert = mcpClient.callTool(
5285 | new CallToolRequest("getAlerts", Map.of("state", "NY")));
5286 |
5287 | mcpClient.closeGracefully();
5288 | ```
5289 |
5290 | ### Use MCP Client Boot Starter
5291 |
5292 | Create a new boot starter applicaiton using the `spring-ai-mcp-client-spring-boot-starter` dependency:
5293 |
5294 | ```xml
5295 | <dependency>
5296 | <groupId>org.springframework.ai</groupId>
5297 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
5298 | </dependency>
5299 | ```
5300 |
5301 | and set the `spring.ai.mcp.client.stdio.servers-configuration` property to point to your `claude_desktop_config.json`.
5302 | You can re-use the existing Anthropic Destop configuration:
5303 |
5304 | ```properties
5305 | spring.ai.mcp.client.stdio.servers-configuration=file:PATH/TO/claude_desktop_config.json
5306 | ```
5307 |
5308 | When you stasrt your client applicaiton, the auto-configuration will create, automatically MCP clients from the claude\_desktop\_config.json.
5309 |
5310 | For more information, see the [MCP Client Boot Starters](https://docs.spring.io/spring-ai/reference/api/mcp/mcp-server-boot-client-docs.html) reference documentation.
5311 |
5312 | ## More Java MCP Server examples
5313 |
5314 | The [starter-webflux-server](https://github.com/spring-projects/spring-ai-examples/tree/main/model-context-protocol/weather/starter-webflux-server) demonstrates how to create a MCP server using SSE transport.
5315 | It showcases how to define and register MCP Tools, Resources, and Prompts, using the Spring Boot's auto-configuration capabilities.
5316 | </Tab>
5317 | </Tabs>
5318 |
5319 | ### Test with commands
5320 |
5321 | Let's make sure Claude for Desktop is picking up the two tools we've exposed in our `weather` server. You can do this by looking for the hammer <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/claude-desktop-mcp-hammer-icon.svg" style={{display: 'inline', margin: 0, height: '1.3em'}} /> icon:
5322 |
5323 | <Frame>
5324 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/visual-indicator-mcp-tools.png" />
5325 | </Frame>
5326 |
5327 | After clicking on the hammer icon, you should see two tools listed:
5328 |
5329 | <Frame>
5330 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/available-mcp-tools.png" />
5331 | </Frame>
5332 |
5333 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
5334 |
5335 | If the hammer icon has shown up, you can now test your server by running the following commands in Claude for Desktop:
5336 |
5337 | * What's the weather in Sacramento?
5338 | * What are the active weather alerts in Texas?
5339 |
5340 | <Frame>
5341 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
5342 | </Frame>
5343 |
5344 | <Frame>
5345 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
5346 | </Frame>
5347 |
5348 | <Note>
5349 | Since this is the US National Weather service, the queries will only work for US locations.
5350 | </Note>
5351 |
5352 | ## What's happening under the hood
5353 |
5354 | When you ask a question:
5355 |
5356 | 1. The client sends your question to Claude
5357 | 2. Claude analyzes the available tools and decides which one(s) to use
5358 | 3. The client executes the chosen tool(s) through the MCP server
5359 | 4. The results are sent back to Claude
5360 | 5. Claude formulates a natural language response
5361 | 6. The response is displayed to you!
5362 |
5363 | ## Troubleshooting
5364 |
5365 | <AccordionGroup>
5366 | <Accordion title="Claude for Desktop Integration Issues">
5367 | **Getting logs from Claude for Desktop**
5368 |
5369 | Claude.app logging related to MCP is written to log files in `~/Library/Logs/Claude`:
5370 |
5371 | * `mcp.log` will contain general logging about MCP connections and connection failures.
5372 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
5373 |
5374 | You can run the following command to list recent logs and follow along with any new ones:
5375 |
5376 | ```bash
5377 | # Check Claude's logs for errors
5378 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
5379 | ```
5380 |
5381 | **Server not showing up in Claude**
5382 |
5383 | 1. Check your `claude_desktop_config.json` file syntax
5384 | 2. Make sure the path to your project is absolute and not relative
5385 | 3. Restart Claude for Desktop completely
5386 |
5387 | **Tool calls failing silently**
5388 |
5389 | If Claude attempts to use the tools but they fail:
5390 |
5391 | 1. Check Claude's logs for errors
5392 | 2. Verify your server builds and runs without errors
5393 | 3. Try restarting Claude for Desktop
5394 |
5395 | **None of this is working. What do I do?**
5396 |
5397 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
5398 | </Accordion>
5399 |
5400 | <Accordion title="Weather API Issues">
5401 | **Error: Failed to retrieve grid point data**
5402 |
5403 | This usually means either:
5404 |
5405 | 1. The coordinates are outside the US
5406 | 2. The NWS API is having issues
5407 | 3. You're being rate limited
5408 |
5409 | Fix:
5410 |
5411 | * Verify you're using US coordinates
5412 | * Add a small delay between requests
5413 | * Check the NWS API status page
5414 |
5415 | **Error: No active alerts for \[STATE]**
5416 |
5417 | This isn't an error - it just means there are no current weather alerts for that state. Try a different state or check during severe weather.
5418 | </Accordion>
5419 | </AccordionGroup>
5420 |
5421 | <Note>
5422 | For more advanced troubleshooting, check out our guide on [Debugging MCP](/docs/tools/debugging)
5423 | </Note>
5424 |
5425 | ## Next steps
5426 |
5427 | <CardGroup cols={2}>
5428 | <Card title="Building a client" icon="outlet" href="/quickstart/client">
5429 | Learn how to build your own MCP client that can connect to your server
5430 | </Card>
5431 |
5432 | <Card title="Example servers" icon="grid" href="/examples">
5433 | Check out our gallery of official MCP servers and implementations
5434 | </Card>
5435 |
5436 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
5437 | Learn how to effectively debug MCP servers and integrations
5438 | </Card>
5439 |
5440 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
5441 | Learn how to use LLMs like Claude to speed up your MCP development
5442 | </Card>
5443 | </CardGroup>
5444 |
5445 |
5446 | # For Claude Desktop Users
5447 | Source: https://modelcontextprotocol.io/quickstart/user
5448 |
5449 | Get started using pre-built servers in Claude for Desktop.
5450 |
5451 | In this tutorial, you will extend [Claude for Desktop](https://claude.ai/download) so that it can read from your computer's file system, write new files, move files, and even search files.
5452 |
5453 | <Frame>
5454 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-filesystem.png" />
5455 | </Frame>
5456 |
5457 | Don't worry — it will ask you for your permission before executing these actions!
5458 |
5459 | ## 1. Download Claude for Desktop
5460 |
5461 | Start by downloading [Claude for Desktop](https://claude.ai/download), choosing either macOS or Windows. (Linux is not yet supported for Claude for Desktop.)
5462 |
5463 | Follow the installation instructions.
5464 |
5465 | If you already have Claude for Desktop, make sure it's on the latest version by clicking on the Claude menu on your computer and selecting "Check for Updates..."
5466 |
5467 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
5468 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
5469 | </Accordion>
5470 |
5471 | ## 2. Add the Filesystem MCP Server
5472 |
5473 | To add this filesystem functionality, we will be installing a pre-built [Filesystem MCP Server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) to Claude for Desktop. This is one of dozens of [servers](https://github.com/modelcontextprotocol/servers/tree/main) created by Anthropic and the community.
5474 |
5475 | Get started by opening up the Claude menu on your computer and select "Settings..." Please note that these are not the Claude Account Settings found in the app window itself.
5476 |
5477 | This is what it should look like on a Mac:
5478 |
5479 | <Frame style={{ textAlign: 'center' }}>
5480 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-menu.png" width="400" />
5481 | </Frame>
5482 |
5483 | Click on "Developer" in the lefthand bar of the Settings pane, and then click on "Edit Config":
5484 |
5485 | <Frame>
5486 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-developer.png" />
5487 | </Frame>
5488 |
5489 | This will create a configuration file at:
5490 |
5491 | * macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
5492 | * Windows: `%APPDATA%\Claude\claude_desktop_config.json`
5493 |
5494 | if you don't already have one, and will display the file in your file system.
5495 |
5496 | Open up the configuration file in any text editor. Replace the file contents with this:
5497 |
5498 | <Tabs>
5499 | <Tab title="MacOS/Linux">
5500 | ```json
5501 | {
5502 | "mcpServers": {
5503 | "filesystem": {
5504 | "command": "npx",
5505 | "args": [
5506 | "-y",
5507 | "@modelcontextprotocol/server-filesystem",
5508 | "/Users/username/Desktop",
5509 | "/Users/username/Downloads"
5510 | ]
5511 | }
5512 | }
5513 | }
5514 | ```
5515 | </Tab>
5516 |
5517 | <Tab title="Windows">
5518 | ```json
5519 | {
5520 | "mcpServers": {
5521 | "filesystem": {
5522 | "command": "npx",
5523 | "args": [
5524 | "-y",
5525 | "@modelcontextprotocol/server-filesystem",
5526 | "C:\\Users\\username\\Desktop",
5527 | "C:\\Users\\username\\Downloads"
5528 | ]
5529 | }
5530 | }
5531 | }
5532 | ```
5533 | </Tab>
5534 | </Tabs>
5535 |
5536 | Make sure to replace `username` with your computer's username. The paths should point to valid directories that you want Claude to be able to access and modify. It's set up to work for Desktop and Downloads, but you can add more paths as well.
5537 |
5538 | You will also need [Node.js](https://nodejs.org) on your computer for this to run properly. To verify you have Node installed, open the command line on your computer.
5539 |
5540 | * On macOS, open the Terminal from your Applications folder
5541 | * On Windows, press Windows + R, type "cmd", and press Enter
5542 |
5543 | Once in the command line, verify you have Node installed by entering in the following command:
5544 |
5545 | ```bash
5546 | node --version
5547 | ```
5548 |
5549 | If you get an error saying "command not found" or "node is not recognized", download Node from [nodejs.org](https://nodejs.org/).
5550 |
5551 | <Tip>
5552 | **How does the configuration file work?**
5553 |
5554 | This configuration file tells Claude for Desktop which MCP servers to start up every time you start the application. In this case, we have added one server called "filesystem" that will use the Node `npx` command to install and run `@modelcontextprotocol/server-filesystem`. This server, described [here](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem), will let you access your file system in Claude for Desktop.
5555 | </Tip>
5556 |
5557 | <Warning>
5558 | **Command Privileges**
5559 |
5560 | Claude for Desktop will run the commands in the configuration file with the permissions of your user account, and access to your local files. Only add commands if you understand and trust the source.
5561 | </Warning>
5562 |
5563 | ## 3. Restart Claude
5564 |
5565 | After updating your configuration file, you need to restart Claude for Desktop.
5566 |
5567 | Upon restarting, you should see a hammer <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/claude-desktop-mcp-hammer-icon.svg" style={{display: 'inline', margin: 0, height: '1.3em'}} /> icon in the bottom right corner of the input box:
5568 |
5569 | <Frame>
5570 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-hammer.png" />
5571 | </Frame>
5572 |
5573 | After clicking on the hammer icon, you should see the tools that come with the Filesystem MCP Server:
5574 |
5575 | <Frame style={{ textAlign: 'center' }}>
5576 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-tools.png" width="400" />
5577 | </Frame>
5578 |
5579 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
5580 |
5581 | ## 4. Try it out!
5582 |
5583 | You can now talk to Claude and ask it about your filesystem. It should know when to call the relevant tools.
5584 |
5585 | Things you might try asking Claude:
5586 |
5587 | * Can you write a poem and save it to my desktop?
5588 | * What are some work-related files in my downloads folder?
5589 | * Can you take all the images on my desktop and move them to a new folder called "Images"?
5590 |
5591 | As needed, Claude will call the relevant tools and seek your approval before taking an action:
5592 |
5593 | <Frame style={{ textAlign: 'center' }}>
5594 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-approve.png" width="500" />
5595 | </Frame>
5596 |
5597 | ## Troubleshooting
5598 |
5599 | <AccordionGroup>
5600 | <Accordion title="Server not showing up in Claude / hammer icon missing">
5601 | 1. Restart Claude for Desktop completely
5602 | 2. Check your `claude_desktop_config.json` file syntax
5603 | 3. Make sure the file paths included in `claude_desktop_config.json` are valid and that they are absolute and not relative
5604 | 4. Look at [logs](#getting-logs-from-claude-for-desktop) to see why the server is not connecting
5605 | 5. In your command line, try manually running the server (replacing `username` as you did in `claude_desktop_config.json`) to see if you get any errors:
5606 |
5607 | <Tabs>
5608 | <Tab title="MacOS/Linux">
5609 | ```bash
5610 | npx -y @modelcontextprotocol/server-filesystem /Users/username/Desktop /Users/username/Downloads
5611 | ```
5612 | </Tab>
5613 |
5614 | <Tab title="Windows">
5615 | ```bash
5616 | npx -y @modelcontextprotocol/server-filesystem C:\Users\username\Desktop C:\Users\username\Downloads
5617 | ```
5618 | </Tab>
5619 | </Tabs>
5620 | </Accordion>
5621 |
5622 | <Accordion title="Getting logs from Claude for Desktop">
5623 | Claude.app logging related to MCP is written to log files in:
5624 |
5625 | * macOS: `~/Library/Logs/Claude`
5626 |
5627 | * Windows: `%APPDATA%\Claude\logs`
5628 |
5629 | * `mcp.log` will contain general logging about MCP connections and connection failures.
5630 |
5631 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
5632 |
5633 | You can run the following command to list recent logs and follow along with any new ones (on Windows, it will only show recent logs):
5634 |
5635 | <Tabs>
5636 | <Tab title="MacOS/Linux">
5637 | ```bash
5638 | # Check Claude's logs for errors
5639 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
5640 | ```
5641 | </Tab>
5642 |
5643 | <Tab title="Windows">
5644 | ```bash
5645 | type "%APPDATA%\Claude\logs\mcp*.log"
5646 | ```
5647 | </Tab>
5648 | </Tabs>
5649 | </Accordion>
5650 |
5651 | <Accordion title="Tool calls failing silently">
5652 | If Claude attempts to use the tools but they fail:
5653 |
5654 | 1. Check Claude's logs for errors
5655 | 2. Verify your server builds and runs without errors
5656 | 3. Try restarting Claude for Desktop
5657 | </Accordion>
5658 |
5659 | <Accordion title="None of this is working. What do I do?">
5660 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
5661 | </Accordion>
5662 |
5663 | <Accordion title="ENOENT error and `${APPDATA}` in paths on Windows">
5664 | If your configured server fails to load, and you see within its logs an error referring to `${APPDATA}` within a path, you may need to add the expanded value of `%APPDATA%` to your `env` key in `claude_desktop_config.json`:
5665 |
5666 | ```json
5667 | {
5668 | "brave-search": {
5669 | "command": "npx",
5670 | "args": ["-y", "@modelcontextprotocol/server-brave-search"],
5671 | "env": {
5672 | "APPDATA": "C:\\Users\\user\\AppData\\Roaming\\",
5673 | "BRAVE_API_KEY": "..."
5674 | }
5675 | }
5676 | }
5677 | ```
5678 |
5679 | With this change in place, launch Claude Desktop once again.
5680 |
5681 | <Warning>
5682 | **NPM should be installed globally**
5683 |
5684 | The `npx` command may continue to fail if you have not installed NPM globally. If NPM is already installed globally, you will find `%APPDATA%\npm` exists on your system. If not, you can install NPM globally by running the following command:
5685 |
5686 | ```bash
5687 | npm install -g npm
5688 | ```
5689 | </Warning>
5690 | </Accordion>
5691 | </AccordionGroup>
5692 |
5693 | ## Next steps
5694 |
5695 | <CardGroup cols={2}>
5696 | <Card title="Explore other servers" icon="grid" href="/examples">
5697 | Check out our gallery of official MCP servers and implementations
5698 | </Card>
5699 |
5700 | <Card title="Build your own server" icon="code" href="/quickstart/server">
5701 | Now build your own custom server to use in Claude for Desktop and other clients
5702 | </Card>
5703 | </CardGroup>
5704 |
5705 |
5706 | # MCP Client
5707 | Source: https://modelcontextprotocol.io/sdk/java/mcp-client
5708 |
5709 | Learn how to use the Model Context Protocol (MCP) client to interact with MCP servers
5710 |
5711 | # Model Context Protocol Client
5712 |
5713 | The MCP Client is a key component in the Model Context Protocol (MCP) architecture, responsible for establishing and managing connections with MCP servers. It implements the client-side of the protocol, handling:
5714 |
5715 | * Protocol version negotiation to ensure compatibility with servers
5716 | * Capability negotiation to determine available features
5717 | * Message transport and JSON-RPC communication
5718 | * Tool discovery and execution
5719 | * Resource access and management
5720 | * Prompt system interactions
5721 | * Optional features like roots management and sampling support
5722 |
5723 | The client provides both synchronous and asynchronous APIs for flexibility in different application contexts.
5724 |
5725 | <Tabs>
5726 | <Tab title="Sync API">
5727 | ```java
5728 | // Create a sync client with custom configuration
5729 | McpSyncClient client = McpClient.sync(transport)
5730 | .requestTimeout(Duration.ofSeconds(10))
5731 | .capabilities(ClientCapabilities.builder()
5732 | .roots(true) // Enable roots capability
5733 | .sampling() // Enable sampling capability
5734 | .build())
5735 | .sampling(request -> new CreateMessageResult(response))
5736 | .build();
5737 |
5738 | // Initialize connection
5739 | client.initialize();
5740 |
5741 | // List available tools
5742 | ListToolsResult tools = client.listTools();
5743 |
5744 | // Call a tool
5745 | CallToolResult result = client.callTool(
5746 | new CallToolRequest("calculator",
5747 | Map.of("operation", "add", "a", 2, "b", 3))
5748 | );
5749 |
5750 | // List and read resources
5751 | ListResourcesResult resources = client.listResources();
5752 | ReadResourceResult resource = client.readResource(
5753 | new ReadResourceRequest("resource://uri")
5754 | );
5755 |
5756 | // List and use prompts
5757 | ListPromptsResult prompts = client.listPrompts();
5758 | GetPromptResult prompt = client.getPrompt(
5759 | new GetPromptRequest("greeting", Map.of("name", "Spring"))
5760 | );
5761 |
5762 | // Add/remove roots
5763 | client.addRoot(new Root("file:///path", "description"));
5764 | client.removeRoot("file:///path");
5765 |
5766 | // Close client
5767 | client.closeGracefully();
5768 | ```
5769 | </Tab>
5770 |
5771 | <Tab title="Async API">
5772 | ```java
5773 | // Create an async client with custom configuration
5774 | McpAsyncClient client = McpClient.async(transport)
5775 | .requestTimeout(Duration.ofSeconds(10))
5776 | .capabilities(ClientCapabilities.builder()
5777 | .roots(true) // Enable roots capability
5778 | .sampling() // Enable sampling capability
5779 | .build())
5780 | .sampling(request -> Mono.just(new CreateMessageResult(response)))
5781 | .toolsChangeConsumer(tools -> Mono.fromRunnable(() -> {
5782 | logger.info("Tools updated: {}", tools);
5783 | }))
5784 | .resourcesChangeConsumer(resources -> Mono.fromRunnable(() -> {
5785 | logger.info("Resources updated: {}", resources);
5786 | }))
5787 | .promptsChangeConsumer(prompts -> Mono.fromRunnable(() -> {
5788 | logger.info("Prompts updated: {}", prompts);
5789 | }))
5790 | .build();
5791 |
5792 | // Initialize connection and use features
5793 | client.initialize()
5794 | .flatMap(initResult -> client.listTools())
5795 | .flatMap(tools -> {
5796 | return client.callTool(new CallToolRequest(
5797 | "calculator",
5798 | Map.of("operation", "add", "a", 2, "b", 3)
5799 | ));
5800 | })
5801 | .flatMap(result -> {
5802 | return client.listResources()
5803 | .flatMap(resources ->
5804 | client.readResource(new ReadResourceRequest("resource://uri"))
5805 | );
5806 | })
5807 | .flatMap(resource -> {
5808 | return client.listPrompts()
5809 | .flatMap(prompts ->
5810 | client.getPrompt(new GetPromptRequest(
5811 | "greeting",
5812 | Map.of("name", "Spring")
5813 | ))
5814 | );
5815 | })
5816 | .flatMap(prompt -> {
5817 | return client.addRoot(new Root("file:///path", "description"))
5818 | .then(client.removeRoot("file:///path"));
5819 | })
5820 | .doFinally(signalType -> {
5821 | client.closeGracefully().subscribe();
5822 | })
5823 | .subscribe();
5824 | ```
5825 | </Tab>
5826 | </Tabs>
5827 |
5828 | ## Client Transport
5829 |
5830 | The transport layer handles the communication between MCP clients and servers, providing different implementations for various use cases. The client transport manages message serialization, connection establishment, and protocol-specific communication patterns.
5831 |
5832 | <Tabs>
5833 | <Tab title="STDIO">
5834 | Creates transport for in-process based communication
5835 |
5836 | ```java
5837 | ServerParameters params = ServerParameters.builder("npx")
5838 | .args("-y", "@modelcontextprotocol/server-everything", "dir")
5839 | .build();
5840 | McpTransport transport = new StdioClientTransport(params);
5841 | ```
5842 | </Tab>
5843 |
5844 | <Tab title="SSE (HttpClient)">
5845 | Creates a framework agnostic (pure Java API) SSE client transport. Included in the core mcp module.
5846 |
5847 | ```java
5848 | McpTransport transport = new HttpClientSseClientTransport("http://your-mcp-server");
5849 | ```
5850 | </Tab>
5851 |
5852 | <Tab title="SSE (WebFlux)">
5853 | Creates WebFlux-based SSE client transport. Requires the mcp-webflux-sse-transport dependency.
5854 |
5855 | ```java
5856 | WebClient.Builder webClientBuilder = WebClient.builder()
5857 | .baseUrl("http://your-mcp-server");
5858 | McpTransport transport = new WebFluxSseClientTransport(webClientBuilder);
5859 | ```
5860 | </Tab>
5861 | </Tabs>
5862 |
5863 | ## Client Capabilities
5864 |
5865 | The client can be configured with various capabilities:
5866 |
5867 | ```java
5868 | var capabilities = ClientCapabilities.builder()
5869 | .roots(true) // Enable filesystem roots support with list changes notifications
5870 | .sampling() // Enable LLM sampling support
5871 | .build();
5872 | ```
5873 |
5874 | ### Roots Support
5875 |
5876 | Roots define the boundaries of where servers can operate within the filesystem:
5877 |
5878 | ```java
5879 | // Add a root dynamically
5880 | client.addRoot(new Root("file:///path", "description"));
5881 |
5882 | // Remove a root
5883 | client.removeRoot("file:///path");
5884 |
5885 | // Notify server of roots changes
5886 | client.rootsListChangedNotification();
5887 | ```
5888 |
5889 | The roots capability allows servers to:
5890 |
5891 | * Request the list of accessible filesystem roots
5892 | * Receive notifications when the roots list changes
5893 | * Understand which directories and files they have access to
5894 |
5895 | ### Sampling Support
5896 |
5897 | Sampling enables servers to request LLM interactions ("completions" or "generations") through the client:
5898 |
5899 | ```java
5900 | // Configure sampling handler
5901 | Function<CreateMessageRequest, CreateMessageResult> samplingHandler = request -> {
5902 | // Sampling implementation that interfaces with LLM
5903 | return new CreateMessageResult(response);
5904 | };
5905 |
5906 | // Create client with sampling support
5907 | var client = McpClient.sync(transport)
5908 | .capabilities(ClientCapabilities.builder()
5909 | .sampling()
5910 | .build())
5911 | .sampling(samplingHandler)
5912 | .build();
5913 | ```
5914 |
5915 | This capability allows:
5916 |
5917 | * Servers to leverage AI capabilities without requiring API keys
5918 | * Clients to maintain control over model access and permissions
5919 | * Support for both text and image-based interactions
5920 | * Optional inclusion of MCP server context in prompts
5921 |
5922 | ## Using MCP Clients
5923 |
5924 | ### Tool Execution
5925 |
5926 | Tools are server-side functions that clients can discover and execute. The MCP client provides methods to list available tools and execute them with specific parameters. Each tool has a unique name and accepts a map of parameters.
5927 |
5928 | <Tabs>
5929 | <Tab title="Sync API">
5930 | ```java
5931 | // List available tools and their names
5932 | var tools = client.listTools();
5933 | tools.forEach(tool -> System.out.println(tool.getName()));
5934 |
5935 | // Execute a tool with parameters
5936 | var result = client.callTool("calculator", Map.of(
5937 | "operation", "add",
5938 | "a", 1,
5939 | "b", 2
5940 | ));
5941 | ```
5942 | </Tab>
5943 |
5944 | <Tab title="Async API">
5945 | ```java
5946 | // List available tools asynchronously
5947 | client.listTools()
5948 | .doOnNext(tools -> tools.forEach(tool ->
5949 | System.out.println(tool.getName())))
5950 | .subscribe();
5951 |
5952 | // Execute a tool asynchronously
5953 | client.callTool("calculator", Map.of(
5954 | "operation", "add",
5955 | "a", 1,
5956 | "b", 2
5957 | ))
5958 | .subscribe();
5959 | ```
5960 | </Tab>
5961 | </Tabs>
5962 |
5963 | ### Resource Access
5964 |
5965 | Resources represent server-side data sources that clients can access using URI templates. The MCP client provides methods to discover available resources and retrieve their contents through a standardized interface.
5966 |
5967 | <Tabs>
5968 | <Tab title="Sync API">
5969 | ```java
5970 | // List available resources and their names
5971 | var resources = client.listResources();
5972 | resources.forEach(resource -> System.out.println(resource.getName()));
5973 |
5974 | // Retrieve resource content using a URI template
5975 | var content = client.getResource("file", Map.of(
5976 | "path", "/path/to/file.txt"
5977 | ));
5978 | ```
5979 | </Tab>
5980 |
5981 | <Tab title="Async API">
5982 | ```java
5983 | // List available resources asynchronously
5984 | client.listResources()
5985 | .doOnNext(resources -> resources.forEach(resource ->
5986 | System.out.println(resource.getName())))
5987 | .subscribe();
5988 |
5989 | // Retrieve resource content asynchronously
5990 | client.getResource("file", Map.of(
5991 | "path", "/path/to/file.txt"
5992 | ))
5993 | .subscribe();
5994 | ```
5995 | </Tab>
5996 | </Tabs>
5997 |
5998 | ### Prompt System
5999 |
6000 | The prompt system enables interaction with server-side prompt templates. These templates can be discovered and executed with custom parameters, allowing for dynamic text generation based on predefined patterns.
6001 |
6002 | <Tabs>
6003 | <Tab title="Sync API">
6004 | ```java
6005 | // List available prompt templates
6006 | var prompts = client.listPrompts();
6007 | prompts.forEach(prompt -> System.out.println(prompt.getName()));
6008 |
6009 | // Execute a prompt template with parameters
6010 | var response = client.executePrompt("echo", Map.of(
6011 | "text", "Hello, World!"
6012 | ));
6013 | ```
6014 | </Tab>
6015 |
6016 | <Tab title="Async API">
6017 | ```java
6018 | // List available prompt templates asynchronously
6019 | client.listPrompts()
6020 | .doOnNext(prompts -> prompts.forEach(prompt ->
6021 | System.out.println(prompt.getName())))
6022 | .subscribe();
6023 |
6024 | // Execute a prompt template asynchronously
6025 | client.executePrompt("echo", Map.of(
6026 | "text", "Hello, World!"
6027 | ))
6028 | .subscribe();
6029 | ```
6030 | </Tab>
6031 | </Tabs>
6032 |
6033 |
6034 | # Overview
6035 | Source: https://modelcontextprotocol.io/sdk/java/mcp-overview
6036 |
6037 | Introduction to the Model Context Protocol (MCP) Java SDK
6038 |
6039 | Java SDK for the [Model Context Protocol](https://modelcontextprotocol.org/docs/concepts/architecture)
6040 | enables standardized integration between AI models and tools.
6041 |
6042 | ## Features
6043 |
6044 | * MCP Client and MCP Server implementations supporting:
6045 | * Protocol [version compatibility negotiation](https://spec.modelcontextprotocol.io/specification/2024-11-05/basic/lifecycle/#initialization)
6046 | * [Tool](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/tools/) discovery, execution, list change notifications
6047 | * [Resource](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/resources/) management with URI templates
6048 | * [Roots](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/roots/) list management and notifications
6049 | * [Prompt](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/prompts/) handling and management
6050 | * [Sampling](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/sampling/) support for AI model interactions
6051 | * Multiple transport implementations:
6052 | * Default transports:
6053 | * Stdio-based transport for process-based communication
6054 | * Java HttpClient-based SSE client transport for HTTP SSE Client-side streaming
6055 | * Servlet-based SSE server transport for HTTP SSE Server streaming
6056 | * Spring-based transports:
6057 | * WebFlux SSE client and server transports for reactive HTTP streaming
6058 | * WebMVC SSE transport for servlet-based HTTP streaming
6059 | * Supports Synchronous and Asynchronous programming paradigms
6060 |
6061 | ## Architecture
6062 |
6063 | The SDK follows a layered architecture with clear separation of concerns:
6064 |
6065 | 
6066 |
6067 | * **Client/Server Layer (McpClient/McpServer)**: Both use McpSession for sync/async operations,
6068 | with McpClient handling client-side protocol operations and McpServer managing server-side protocol operations.
6069 | * **Session Layer (McpSession)**: Manages communication patterns and state using DefaultMcpSession implementation.
6070 | * **Transport Layer (McpTransport)**: Handles JSON-RPC message serialization/deserialization via:
6071 | * StdioTransport (stdin/stdout) in the core module
6072 | * HTTP SSE transports in dedicated transport modules (Java HttpClient, Spring WebFlux, Spring WebMVC)
6073 |
6074 | The MCP Client is a key component in the Model Context Protocol (MCP) architecture, responsible for establishing and managing connections with MCP servers.
6075 | It implements the client-side of the protocol.
6076 |
6077 | 
6078 |
6079 | The MCP Server is a foundational component in the Model Context Protocol (MCP) architecture that provides tools, resources, and capabilities to clients.
6080 | It implements the server-side of the protocol.
6081 |
6082 | 
6083 |
6084 | Key Interactions:
6085 |
6086 | * **Client/Server Initialization**: Transport setup, protocol compatibility check, capability negotiation, and implementation details exchange.
6087 | * **Message Flow**: JSON-RPC message handling with validation, type-safe response processing, and error handling.
6088 | * **Resource Management**: Resource discovery, URI template-based access, subscription system, and content retrieval.
6089 |
6090 | ## Dependencies
6091 |
6092 | Add the following Maven dependency to your project:
6093 |
6094 | <Tabs>
6095 | <Tab title="Maven">
6096 | The core MCP functionality:
6097 |
6098 | ```xml
6099 | <dependency>
6100 | <groupId>io.modelcontextprotocol.sdk</groupId>
6101 | <artifactId>mcp</artifactId>
6102 | </dependency>
6103 | ```
6104 |
6105 | For HTTP SSE transport implementations, add one of the following dependencies:
6106 |
6107 | ```xml
6108 | <!-- Spring WebFlux-based SSE client and server transport -->
6109 | <dependency>
6110 | <groupId>io.modelcontextprotocol.sdk</groupId>
6111 | <artifactId>mcp-spring-webflux</artifactId>
6112 | </dependency>
6113 |
6114 | <!-- Spring WebMVC-based SSE server transport -->
6115 | <dependency>
6116 | <groupId>io.modelcontextprotocol.sdk</groupId>
6117 | <artifactId>mcp-spring-webmvc</artifactId>
6118 | </dependency>
6119 | ```
6120 | </Tab>
6121 |
6122 | <Tab title="Gradle">
6123 | The core MCP functionality:
6124 |
6125 | ```groovy
6126 | dependencies {
6127 | implementation platform("io.modelcontextprotocol.sdk:mcp")
6128 | //...
6129 | }
6130 | ```
6131 |
6132 | For HTTP SSE transport implementations, add one of the following dependencies:
6133 |
6134 | ```groovy
6135 | // Spring WebFlux-based SSE client and server transport
6136 | dependencies {
6137 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webflux")
6138 | }
6139 |
6140 | // Spring WebMVC-based SSE server transport
6141 | dependencies {
6142 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webmvc")
6143 | }
6144 | ```
6145 | </Tab>
6146 | </Tabs>
6147 |
6148 | ### Bill of Materials (BOM)
6149 |
6150 | The Bill of Materials (BOM) declares the recommended versions of all the dependencies used by a given release.
6151 | Using the BOM from your application's build script avoids the need for you to specify and maintain the dependency versions yourself.
6152 | Instead, the version of the BOM you're using determines the utilized dependency versions.
6153 | It also ensures that you're using supported and tested versions of the dependencies by default, unless you choose to override them.
6154 |
6155 | Add the BOM to your project:
6156 |
6157 | <Tabs>
6158 | <Tab title="Maven">
6159 | ```xml
6160 | <dependencyManagement>
6161 | <dependencies>
6162 | <dependency>
6163 | <groupId>io.modelcontextprotocol.sdk</groupId>
6164 | <artifactId>mcp-bom</artifactId>
6165 | <version>0.7.0</version>
6166 | <type>pom</type>
6167 | <scope>import</scope>
6168 | </dependency>
6169 | </dependencies>
6170 | </dependencyManagement>
6171 | ```
6172 | </Tab>
6173 |
6174 | <Tab title="Gradle">
6175 | ```groovy
6176 | dependencies {
6177 | implementation platform("io.modelcontextprotocol.sdk:mcp-bom:0.7.0")
6178 | //...
6179 | }
6180 | ```
6181 |
6182 | Gradle users can also use the Spring AI MCP BOM by leveraging Gradle (5.0+) native support for declaring dependency constraints using a Maven BOM.
6183 | This is implemented by adding a 'platform' dependency handler method to the dependencies section of your Gradle build script.
6184 | As shown in the snippet above this can then be followed by version-less declarations of the Starter Dependencies for the one or more spring-ai modules you wish to use, e.g. spring-ai-openai.
6185 | </Tab>
6186 | </Tabs>
6187 |
6188 | Replace the version number with the version of the BOM you want to use.
6189 |
6190 | ### Available Dependencies
6191 |
6192 | The following dependencies are available and managed by the BOM:
6193 |
6194 | * Core Dependencies
6195 | * `io.modelcontextprotocol.sdk:mcp` - Core MCP library providing the base functionality and APIs for Model Context Protocol implementation.
6196 | * Transport Dependencies
6197 | * `io.modelcontextprotocol.sdk:mcp-spring-webflux` - WebFlux-based Server-Sent Events (SSE) transport implementation for reactive applications.
6198 | * `io.modelcontextprotocol.sdk:mcp-spring-webmvc` - WebMVC-based Server-Sent Events (SSE) transport implementation for servlet-based applications.
6199 | * Testing Dependencies
6200 | * `io.modelcontextprotocol.sdk:mcp-test` - Testing utilities and support for MCP-based applications.
6201 |
6202 |
6203 | # MCP Server
6204 | Source: https://modelcontextprotocol.io/sdk/java/mcp-server
6205 |
6206 | Learn how to implement and configure a Model Context Protocol (MCP) server
6207 |
6208 | ## Overview
6209 |
6210 | The MCP Server is a foundational component in the Model Context Protocol (MCP) architecture that provides tools, resources, and capabilities to clients. It implements the server-side of the protocol, responsible for:
6211 |
6212 | * Exposing tools that clients can discover and execute
6213 | * Managing resources with URI-based access patterns
6214 | * Providing prompt templates and handling prompt requests
6215 | * Supporting capability negotiation with clients
6216 | * Implementing server-side protocol operations
6217 | * Managing concurrent client connections
6218 | * Providing structured logging and notifications
6219 |
6220 | The server supports both synchronous and asynchronous APIs, allowing for flexible integration in different application contexts.
6221 |
6222 | <Tabs>
6223 | <Tab title="Sync API">
6224 | ```java
6225 | // Create a server with custom configuration
6226 | McpSyncServer syncServer = McpServer.sync(transport)
6227 | .serverInfo("my-server", "1.0.0")
6228 | .capabilities(ServerCapabilities.builder()
6229 | .resources(true) // Enable resource support
6230 | .tools(true) // Enable tool support
6231 | .prompts(true) // Enable prompt support
6232 | .logging() // Enable logging support
6233 | .build())
6234 | .build();
6235 |
6236 | // Register tools, resources, and prompts
6237 | syncServer.addTool(syncToolRegistration);
6238 | syncServer.addResource(syncResourceRegistration);
6239 | syncServer.addPrompt(syncPromptRegistration);
6240 |
6241 | // Send logging notifications
6242 | syncServer.loggingNotification(LoggingMessageNotification.builder()
6243 | .level(LoggingLevel.INFO)
6244 | .logger("custom-logger")
6245 | .data("Server initialized")
6246 | .build());
6247 |
6248 | // Close the server when done
6249 | syncServer.close();
6250 | ```
6251 | </Tab>
6252 |
6253 | <Tab title="Async API">
6254 | ```java
6255 | // Create an async server with custom configuration
6256 | McpAsyncServer asyncServer = McpServer.async(transport)
6257 | .serverInfo("my-server", "1.0.0")
6258 | .capabilities(ServerCapabilities.builder()
6259 | .resources(true) // Enable resource support
6260 | .tools(true) // Enable tool support
6261 | .prompts(true) // Enable prompt support
6262 | .logging() // Enable logging support
6263 | .build())
6264 | .build();
6265 |
6266 | // Register tools, resources, and prompts
6267 | asyncServer.addTool(asyncToolRegistration)
6268 | .doOnSuccess(v -> logger.info("Tool registered"))
6269 | .subscribe();
6270 |
6271 | asyncServer.addResource(asyncResourceRegistration)
6272 | .doOnSuccess(v -> logger.info("Resource registered"))
6273 | .subscribe();
6274 |
6275 | asyncServer.addPrompt(asyncPromptRegistration)
6276 | .doOnSuccess(v -> logger.info("Prompt registered"))
6277 | .subscribe();
6278 |
6279 | // Send logging notifications
6280 | asyncServer.loggingNotification(LoggingMessageNotification.builder()
6281 | .level(LoggingLevel.INFO)
6282 | .logger("custom-logger")
6283 | .data("Server initialized")
6284 | .build());
6285 |
6286 | // Close the server when done
6287 | asyncServer.close()
6288 | .doOnSuccess(v -> logger.info("Server closed"))
6289 | .subscribe();
6290 | ```
6291 | </Tab>
6292 | </Tabs>
6293 |
6294 | ## Server Transport
6295 |
6296 | The transport layer in the MCP SDK is responsible for handling the communication between clients and servers. It provides different implementations to support various communication protocols and patterns. The SDK includes several built-in transport implementations:
6297 |
6298 | <Tabs>
6299 | <Tab title="STDIO">
6300 | <>
6301 | Create in-process based transport:
6302 |
6303 | ```java
6304 | StdioServerTransport transport = new StdioServerTransport(new ObjectMapper());
6305 | ```
6306 |
6307 | Provides bidirectional JSON-RPC message handling over standard input/output streams with non-blocking message processing, serialization/deserialization, and graceful shutdown support.
6308 |
6309 | Key features:
6310 |
6311 | <ul>
6312 | <li>Bidirectional communication through stdin/stdout</li>
6313 | <li>Process-based integration support</li>
6314 | <li>Simple setup and configuration</li>
6315 | <li>Lightweight implementation</li>
6316 | </ul>
6317 | </>
6318 | </Tab>
6319 |
6320 | <Tab title="SSE (WebFlux)">
6321 | <>
6322 | <p>Creates WebFlux-based SSE server transport.<br />Requires the <code>mcp-spring-webflux</code> dependency.</p>
6323 |
6324 | ```java
6325 | @Configuration
6326 | class McpConfig {
6327 | @Bean
6328 | WebFluxSseServerTransport webFluxSseServerTransport(ObjectMapper mapper) {
6329 | return new WebFluxSseServerTransport(mapper, "/mcp/message");
6330 | }
6331 |
6332 | @Bean
6333 | RouterFunction<?> mcpRouterFunction(WebFluxSseServerTransport transport) {
6334 | return transport.getRouterFunction();
6335 | }
6336 | }
6337 | ```
6338 |
6339 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
6340 |
6341 | <ul>
6342 | <li>Reactive HTTP streaming with WebFlux</li>
6343 | <li>Concurrent client connections through SSE endpoints</li>
6344 | <li>Message routing and session management</li>
6345 | <li>Graceful shutdown capabilities</li>
6346 | </ul>
6347 | </>
6348 | </Tab>
6349 |
6350 | <Tab title="SSE (WebMvc)">
6351 | <>
6352 | <p>Creates WebMvc-based SSE server transport.<br />Requires the <code>mcp-spring-webmvc</code> dependency.</p>
6353 |
6354 | ```java
6355 | @Configuration
6356 | @EnableWebMvc
6357 | class McpConfig {
6358 | @Bean
6359 | WebMvcSseServerTransport webMvcSseServerTransport(ObjectMapper mapper) {
6360 | return new WebMvcSseServerTransport(mapper, "/mcp/message");
6361 | }
6362 |
6363 | @Bean
6364 | RouterFunction<ServerResponse> mcpRouterFunction(WebMvcSseServerTransport transport) {
6365 | return transport.getRouterFunction();
6366 | }
6367 | }
6368 | ```
6369 |
6370 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
6371 |
6372 | <ul>
6373 | <li>Server-side event streaming</li>
6374 | <li>Integration with Spring WebMVC</li>
6375 | <li>Support for traditional web applications</li>
6376 | <li>Synchronous operation handling</li>
6377 | </ul>
6378 | </>
6379 | </Tab>
6380 |
6381 | <Tab title="SSE (Servlet)">
6382 | <>
6383 | <p>
6384 | Creates a Servlet-based SSE server transport. It is included in the core <code>mcp</code> module.<br />
6385 | The <code>HttpServletSseServerTransport</code> can be used with any Servlet container.<br />
6386 | To use it with a Spring Web application, you can register it as a Servlet bean:
6387 | </p>
6388 |
6389 | ```java
6390 | @Configuration
6391 | @EnableWebMvc
6392 | public class McpServerConfig implements WebMvcConfigurer {
6393 |
6394 | @Bean
6395 | public HttpServletSseServerTransport servletSseServerTransport() {
6396 | return new HttpServletSseServerTransport(new ObjectMapper(), "/mcp/message");
6397 | }
6398 |
6399 | @Bean
6400 | public ServletRegistrationBean customServletBean(HttpServletSseServerTransport servlet) {
6401 | return new ServletRegistrationBean(servlet);
6402 | }
6403 | }
6404 | ```
6405 |
6406 | <p>
6407 | Implements the MCP HTTP with SSE transport specification using the traditional Servlet API, providing:
6408 | </p>
6409 |
6410 | <ul>
6411 | <li>Asynchronous message handling using Servlet 6.0 async support</li>
6412 | <li>Session management for multiple client connections</li>
6413 |
6414 | <li>
6415 | Two types of endpoints:
6416 |
6417 | <ul>
6418 | <li>SSE endpoint (<code>/sse</code>) for server-to-client events</li>
6419 | <li>Message endpoint (configurable) for client-to-server requests</li>
6420 | </ul>
6421 | </li>
6422 |
6423 | <li>Error handling and response formatting</li>
6424 | <li>Graceful shutdown support</li>
6425 | </ul>
6426 | </>
6427 | </Tab>
6428 | </Tabs>
6429 |
6430 | ## Server Capabilities
6431 |
6432 | The server can be configured with various capabilities:
6433 |
6434 | ```java
6435 | var capabilities = ServerCapabilities.builder()
6436 | .resources(false, true) // Resource support with list changes notifications
6437 | .tools(true) // Tool support with list changes notifications
6438 | .prompts(true) // Prompt support with list changes notifications
6439 | .logging() // Enable logging support (enabled by default with loging level INFO)
6440 | .build();
6441 | ```
6442 |
6443 | ### Logging Support
6444 |
6445 | The server provides structured logging capabilities that allow sending log messages to clients with different severity levels:
6446 |
6447 | ```java
6448 | // Send a log message to clients
6449 | server.loggingNotification(LoggingMessageNotification.builder()
6450 | .level(LoggingLevel.INFO)
6451 | .logger("custom-logger")
6452 | .data("Custom log message")
6453 | .build());
6454 | ```
6455 |
6456 | Clients can control the minimum logging level they receive through the `mcpClient.setLoggingLevel(level)` request. Messages below the set level will be filtered out.
6457 | Supported logging levels (in order of increasing severity): DEBUG (0), INFO (1), NOTICE (2), WARNING (3), ERROR (4), CRITICAL (5), ALERT (6), EMERGENCY (7)
6458 |
6459 | ### Tool Registration
6460 |
6461 | <Tabs>
6462 | <Tab title="Sync">
6463 | ```java
6464 | // Sync tool registration
6465 | var schema = """
6466 | {
6467 | "type" : "object",
6468 | "id" : "urn:jsonschema:Operation",
6469 | "properties" : {
6470 | "operation" : {
6471 | "type" : "string"
6472 | },
6473 | "a" : {
6474 | "type" : "number"
6475 | },
6476 | "b" : {
6477 | "type" : "number"
6478 | }
6479 | }
6480 | }
6481 | """;
6482 | var syncToolRegistration = new McpServerFeatures.SyncToolRegistration(
6483 | new Tool("calculator", "Basic calculator", schema),
6484 | arguments -> {
6485 | // Tool implementation
6486 | return new CallToolResult(result, false);
6487 | }
6488 | );
6489 | ```
6490 | </Tab>
6491 |
6492 | <Tab title="Async">
6493 | ```java
6494 | // Async tool registration
6495 | var schema = """
6496 | {
6497 | "type" : "object",
6498 | "id" : "urn:jsonschema:Operation",
6499 | "properties" : {
6500 | "operation" : {
6501 | "type" : "string"
6502 | },
6503 | "a" : {
6504 | "type" : "number"
6505 | },
6506 | "b" : {
6507 | "type" : "number"
6508 | }
6509 | }
6510 | }
6511 | """;
6512 | var asyncToolRegistration = new McpServerFeatures.AsyncToolRegistration(
6513 | new Tool("calculator", "Basic calculator", schema),
6514 | arguments -> {
6515 | // Tool implementation
6516 | return Mono.just(new CallToolResult(result, false));
6517 | }
6518 | );
6519 | ```
6520 | </Tab>
6521 | </Tabs>
6522 |
6523 | ### Resource Registration
6524 |
6525 | <Tabs>
6526 | <Tab title="Sync">
6527 | ```java
6528 | // Sync resource registration
6529 | var syncResourceRegistration = new McpServerFeatures.SyncResourceRegistration(
6530 | new Resource("custom://resource", "name", "description", "mime-type", null),
6531 | request -> {
6532 | // Resource read implementation
6533 | return new ReadResourceResult(contents);
6534 | }
6535 | );
6536 | ```
6537 | </Tab>
6538 |
6539 | <Tab title="Async">
6540 | ```java
6541 | // Async resource registration
6542 | var asyncResourceRegistration = new McpServerFeatures.AsyncResourceRegistration(
6543 | new Resource("custom://resource", "name", "description", "mime-type", null),
6544 | request -> {
6545 | // Resource read implementation
6546 | return Mono.just(new ReadResourceResult(contents));
6547 | }
6548 | );
6549 | ```
6550 | </Tab>
6551 | </Tabs>
6552 |
6553 | ### Prompt Registration
6554 |
6555 | <Tabs>
6556 | <Tab title="Sync">
6557 | ```java
6558 | // Sync prompt registration
6559 | var syncPromptRegistration = new McpServerFeatures.SyncPromptRegistration(
6560 | new Prompt("greeting", "description", List.of(
6561 | new PromptArgument("name", "description", true)
6562 | )),
6563 | request -> {
6564 | // Prompt implementation
6565 | return new GetPromptResult(description, messages);
6566 | }
6567 | );
6568 | ```
6569 | </Tab>
6570 |
6571 | <Tab title="Async">
6572 | ```java
6573 | // Async prompt registration
6574 | var asyncPromptRegistration = new McpServerFeatures.AsyncPromptRegistration(
6575 | new Prompt("greeting", "description", List.of(
6576 | new PromptArgument("name", "description", true)
6577 | )),
6578 | request -> {
6579 | // Prompt implementation
6580 | return Mono.just(new GetPromptResult(description, messages));
6581 | }
6582 | );
6583 | ```
6584 | </Tab>
6585 | </Tabs>
6586 |
6587 | ## Error Handling
6588 |
6589 | The SDK provides comprehensive error handling through the McpError class, covering protocol compatibility, transport communication, JSON-RPC messaging, tool execution, resource management, prompt handling, timeouts, and connection issues. This unified error handling approach ensures consistent and reliable error management across both synchronous and asynchronous operations.
6590 |
6591 |
6592 | # Building MCP with LLMs
6593 | Source: https://modelcontextprotocol.io/tutorials/building-mcp-with-llms
6594 |
6595 | Speed up your MCP development using LLMs such as Claude!
6596 |
6597 | This guide will help you use LLMs to help you build custom Model Context Protocol (MCP) servers and clients. We'll be focusing on Claude for this tutorial, but you can do this with any frontier LLM.
6598 |
6599 | ## Preparing the documentation
6600 |
6601 | Before starting, gather the necessary documentation to help Claude understand MCP:
6602 |
6603 | 1. Visit [https://modelcontextprotocol.io/llms-full.txt](https://modelcontextprotocol.io/llms-full.txt) and copy the full documentation text
6604 | 2. Navigate to either the [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) or [Python SDK repository](https://github.com/modelcontextprotocol/python-sdk)
6605 | 3. Copy the README files and other relevant documentation
6606 | 4. Paste these documents into your conversation with Claude
6607 |
6608 | ## Describing your server
6609 |
6610 | Once you've provided the documentation, clearly describe to Claude what kind of server you want to build. Be specific about:
6611 |
6612 | * What resources your server will expose
6613 | * What tools it will provide
6614 | * Any prompts it should offer
6615 | * What external systems it needs to interact with
6616 |
6617 | For example:
6618 |
6619 | ```
6620 | Build an MCP server that:
6621 | - Connects to my company's PostgreSQL database
6622 | - Exposes table schemas as resources
6623 | - Provides tools for running read-only SQL queries
6624 | - Includes prompts for common data analysis tasks
6625 | ```
6626 |
6627 | ## Working with Claude
6628 |
6629 | When working with Claude on MCP servers:
6630 |
6631 | 1. Start with the core functionality first, then iterate to add more features
6632 | 2. Ask Claude to explain any parts of the code you don't understand
6633 | 3. Request modifications or improvements as needed
6634 | 4. Have Claude help you test the server and handle edge cases
6635 |
6636 | Claude can help implement all the key MCP features:
6637 |
6638 | * Resource management and exposure
6639 | * Tool definitions and implementations
6640 | * Prompt templates and handlers
6641 | * Error handling and logging
6642 | * Connection and transport setup
6643 |
6644 | ## Best practices
6645 |
6646 | When building MCP servers with Claude:
6647 |
6648 | * Break down complex servers into smaller pieces
6649 | * Test each component thoroughly before moving on
6650 | * Keep security in mind - validate inputs and limit access appropriately
6651 | * Document your code well for future maintenance
6652 | * Follow MCP protocol specifications carefully
6653 |
6654 | ## Next steps
6655 |
6656 | After Claude helps you build your server:
6657 |
6658 | 1. Review the generated code carefully
6659 | 2. Test the server with the MCP Inspector tool
6660 | 3. Connect it to Claude.app or other MCP clients
6661 | 4. Iterate based on real usage and feedback
6662 |
6663 | Remember that Claude can help you modify and improve your server as requirements change over time.
6664 |
6665 | Need more guidance? Just ask Claude specific questions about implementing MCP features or troubleshooting issues that arise.
6666 |
6667 |
```