This is page 5 of 5. Use http://codebase.md/stumason/coolify-mcp?lines=true&page={x} to view the full context.
# Directory Structure
```
├── .cursor
│ └── rules
│ ├── 000-cursor-rules.mdc
│ ├── 801-feature-workflow.mdc
│ ├── 802-coolify-mcp-workflow.mdc
│ └── 803-npm-publish-workflow.mdc
├── .eslintrc.json
├── .github
│ └── workflows
│ └── ci.yml
├── .gitignore
├── .lintstagedrc.json
├── .markdownlint-cli2.jsonc
├── .prettierrc
├── .repomixignore
├── debug.js
├── docs
│ ├── coolify-openapi.yaml
│ ├── features
│ │ ├── 001-core-server-setup.md
│ │ ├── 002-server-info-resource.md
│ │ ├── 003-project-management.md
│ │ ├── 004-environment-management.md
│ │ ├── 005-application-deployment.md
│ │ ├── 006-database-management.md
│ │ ├── 007-service-management.md
│ │ ├── 008-mcp-resources-implementation.md
│ │ ├── 009-mcp-prompts-implementation.md
│ │ ├── 010-private-key-management.md
│ │ ├── 011-team-management.md
│ │ ├── 012-backup-management.md
│ │ ├── 013-npx-config-fix.md
│ │ └── future-adrs.md
│ ├── mcp-example-clients.md
│ ├── mcp-js-readme.md
│ └── openapi-chunks
│ ├── applications-api.yaml
│ ├── databases-api.yaml
│ ├── deployments-api.yaml
│ ├── private-keys-api.yaml
│ ├── projects-api.yaml
│ ├── resources-api.yaml
│ ├── schemas.yaml
│ ├── servers-api.yaml
│ ├── services-api.yaml
│ ├── teams-api.yaml
│ └── untagged-api.yaml
├── jest.config.js
├── package-lock.json
├── package.json
├── README.md
├── repomix-output.xml
├── src
│ ├── __tests__
│ │ ├── coolify-client.test.ts
│ │ └── resources
│ │ ├── application-resources.test.ts
│ │ ├── database-resources.test.ts
│ │ ├── deployment-resources.test.ts
│ │ └── service-resources.test.ts
│ ├── index.ts
│ ├── lib
│ │ ├── coolify-client.ts
│ │ ├── mcp-server.ts
│ │ └── resource.ts
│ ├── resources
│ │ ├── application-resources.ts
│ │ ├── database-resources.ts
│ │ ├── deployment-resources.ts
│ │ ├── index.ts
│ │ └── service-resources.ts
│ └── types
│ └── coolify.ts
└── tsconfig.json
```
# Files
--------------------------------------------------------------------------------
/docs/mcp-example-clients.md:
--------------------------------------------------------------------------------
```markdown
1 | # Example Clients
2 |
3 | Source: https://modelcontextprotocol.io/clients
4 |
5 | A list of applications that support MCP integrations
6 |
7 | 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.
8 |
9 | ## Feature support matrix
10 |
11 | | Client | [Resources] | [Prompts] | [Tools] | [Sampling] | Roots | Notes |
12 | | ------------------------------------ | ----------- | --------- | ------- | ---------- | ----- | ------------------------------------------------------------------ |
13 | | [Claude Desktop App][Claude] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
14 | | [5ire][5ire] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
15 | | [BeeAI Framework][BeeAI Framework] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in agentic workflows. |
16 | | [Cline][Cline] | ✅ | ❌ | ✅ | ❌ | ❌ | Supports tools and resources. |
17 | | [Continue][Continue] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
18 | | [Cursor][Cursor] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
19 | | [Emacs Mcp][Mcp.el] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in Emacs. |
20 | | [Firebase Genkit][Genkit] | ⚠️ | ✅ | ✅ | ❌ | ❌ | Supports resource list and lookup through tools. |
21 | | [GenAIScript][GenAIScript] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
22 | | [Goose][Goose] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
23 | | [LibreChat][LibreChat] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Agents |
24 | | [mcp-agent][mcp-agent] | ❌ | ❌ | ✅ | ⚠️ | ❌ | Supports tools, server connection management, and agent workflows. |
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 | | \[OpenSumi]\[OpenSumi] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in OpenSumi |
32 |
33 | [Claude]: https://claude.ai/download
34 | [Cursor]: https://cursor.com
35 | [Zed]: https://zed.dev
36 | [Cody]: https://sourcegraph.com/cody
37 | [Genkit]: https://github.com/firebase/genkit
38 | [Continue]: https://github.com/continuedev/continue
39 | [GenAIScript]: https://microsoft.github.io/genaiscript/reference/scripts/mcp-tools/
40 | [Cline]: https://github.com/cline/cline
41 | [LibreChat]: https://github.com/danny-avila/LibreChat
42 | [TheiaAI/TheiaIDE]: https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/
43 | [Superinterface]: https://superinterface.ai
44 | [5ire]: https://github.com/nanbingxyz/5ire
45 | [BeeAI Framework]: https://i-am-bee.github.io/beeai-framework
46 | [mcp-agent]: https://github.com/lastmile-ai/mcp-agent
47 | [Mcp.el]: https://github.com/lizqwerscott/mcp.el
48 | [Roo Code]: https://roocode.com
49 | [Goose]: https://block.github.io/goose/docs/goose-architecture/#interoperability-with-extensions
50 | [Windsurf]: https://codeium.com/windsurf
51 | [Resources]: https://modelcontextprotocol.io/docs/concepts/resources
52 | [Prompts]: https://modelcontextprotocol.io/docs/concepts/prompts
53 | [Tools]: https://modelcontextprotocol.io/docs/concepts/tools
54 | [Sampling]: https://modelcontextprotocol.io/docs/concepts/sampling
55 |
56 | ## Client details
57 |
58 | ### Claude Desktop App
59 |
60 | The Claude desktop application provides comprehensive support for MCP, enabling deep integration with local tools and data sources.
61 |
62 | **Key features:**
63 |
64 | - Full support for resources, allowing attachment of local files and data
65 | - Support for prompt templates
66 | - Tool integration for executing commands and scripts
67 | - Local server connections for enhanced privacy and security
68 |
69 | > ⓘ Note: The Claude.ai web application does not currently support MCP. MCP features are only available in the desktop application.
70 |
71 | ### 5ire
72 |
73 | [5ire](https://github.com/nanbingxyz/5ire) is an open source cross-platform desktop AI assistant that supports tools through MCP servers.
74 |
75 | **Key features:**
76 |
77 | - Built-in MCP servers can be quickly enabled and disabled.
78 | - Users can add more servers by modifying the configuration file.
79 | - It is open-source and user-friendly, suitable for beginners.
80 | - Future support for MCP will be continuously improved.
81 |
82 | ### BeeAI Framework
83 |
84 | [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.
85 |
86 | **Key features:**
87 |
88 | - Seamlessly incorporate MCP tools into agentic workflows.
89 | - Quickly instantiate framework-native tools from connected MCP client(s).
90 | - Planned future support for agentic MCP capabilities.
91 |
92 | **Learn more:**
93 |
94 | - [Example of using MCP tools in agentic workflow](https://i-am-bee.github.io/beeai-framework/#/typescript/tools?id=using-the-mcptool-class)
95 |
96 | ### Cline
97 |
98 | [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.
99 |
100 | **Key features:**
101 |
102 | - Create and add tools through natural language (e.g. "add a tool that searches the web")
103 | - Share custom MCP servers Cline creates with others via the `~/Documents/Cline/MCP` directory
104 | - Displays configured MCP servers along with their tools, resources, and any error logs
105 |
106 | ### Continue
107 |
108 | [Continue](https://github.com/continuedev/continue) is an open-source AI code assistant, with built-in support for all MCP features.
109 |
110 | **Key features**
111 |
112 | - Type "@" to mention MCP resources
113 | - Prompt templates surface as slash commands
114 | - Use both built-in and MCP tools directly in chat
115 | - Supports VS Code and JetBrains IDEs, with any LLM
116 |
117 | ### Cursor
118 |
119 | [Cursor](https://docs.cursor.com/advanced/model-context-protocol) is an AI code editor.
120 |
121 | **Key Features**:
122 |
123 | - Support for MCP tools in Cursor Composer
124 | - Support for both STDIO and SSE
125 |
126 | ### Emacs Mcp
127 |
128 | [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.
129 |
130 | **Key features:**
131 |
132 | - Provides MCP tool support for Emacs.
133 |
134 | ### Firebase Genkit
135 |
136 | [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.
137 |
138 | **Key features:**
139 |
140 | - Client support for tools and prompts (resources partially supported)
141 | - Rich discovery with support in Genkit's Dev UI playground
142 | - Seamless interoperability with Genkit's existing tools and prompts
143 | - Works across a wide variety of GenAI models from top providers
144 |
145 | ### GenAIScript
146 |
147 | Programmatically assemble prompts for LLMs using [GenAIScript](https://microsoft.github.io/genaiscript/) (in JavaScript). Orchestrate LLMs, tools, and data in JavaScript.
148 |
149 | **Key features:**
150 |
151 | - JavaScript toolbox to work with prompts
152 | - Abstraction to make it easy and productive
153 | - Seamless Visual Studio Code integration
154 |
155 | ### Goose
156 |
157 | [Goose](https://github.com/block/goose) is an open source AI agent that supercharges your software development by automating coding tasks.
158 |
159 | **Key features:**
160 |
161 | - Expose MCP functionality to Goose through tools.
162 | - MCPs can be installed directly via the [extensions directory](https://block.github.io/goose/v1/extensions/), CLI, or UI.
163 | - Goose allows you to extend its functionality by [building your own MCP servers](https://block.github.io/goose/docs/tutorials/custom-extensions).
164 | - Includes built-in tools for development, web scraping, automation, memory, and integrations with JetBrains and Google Drive.
165 |
166 | ### LibreChat
167 |
168 | [LibreChat](https://github.com/danny-avila/LibreChat) is an open-source, customizable AI chat UI that supports multiple AI providers, now including MCP integration.
169 |
170 | **Key features:**
171 |
172 | - Extend current tool ecosystem, including [Code Interpreter](https://www.librechat.ai/docs/features/code_interpreter) and Image generation tools, through MCP servers
173 | - Add tools to customizable [Agents](https://www.librechat.ai/docs/features/agents), using a variety of LLMs from top providers
174 | - Open-source and self-hostable, with secure multi-user support
175 | - Future roadmap includes expanded MCP feature support
176 |
177 | ### mcp-agent
178 |
179 | [mcp-agent] is a simple, composable framework to build agents using Model Context Protocol.
180 |
181 | **Key features:**
182 |
183 | - Automatic connection management of MCP servers.
184 | - Expose tools from multiple servers to an LLM.
185 | - Implements every pattern defined in [Building Effective Agents](https://www.anthropic.com/research/building-effective-agents).
186 | - Supports workflow pause/resume signals, such as waiting for human feedback.
187 |
188 | ### Roo Code
189 |
190 | [Roo Code](https://roocode.com) enables AI coding assistance via MCP.
191 |
192 | **Key features:**
193 |
194 | - Support for MCP tools and resources
195 | - Integration with development workflows
196 | - Extensible AI capabilities
197 |
198 | ### Sourcegraph Cody
199 |
200 | [Cody](https://openctx.org/docs/providers/modelcontextprotocol) is Sourcegraph's AI coding assistant, which implements MCP through OpenCTX.
201 |
202 | **Key features:**
203 |
204 | - Support for MCP resources
205 | - Integration with Sourcegraph's code intelligence
206 | - Uses OpenCTX as an abstraction layer
207 | - Future support planned for additional MCP features
208 |
209 | ### Superinterface
210 |
211 | [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.
212 |
213 | **Key features:**
214 |
215 | - Use tools from MCP servers in assistants embedded via React components or script tags
216 | - SSE transport support
217 | - Use any AI model from any AI provider (OpenAI, Anthropic, Ollama, others)
218 |
219 | ### TheiaAI/TheiaIDE
220 |
221 | [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.
222 |
223 | **Key features:**
224 |
225 | - **Tool Integration**: Theia AI enables AI agents, including those in the Theia IDE, to utilize MCP servers for seamless tool interaction.
226 | - **Customizable Prompts**: The Theia IDE allows users to define and adapt prompts, dynamically integrating MCP servers for tailored workflows.
227 | - **Custom agents**: The Theia IDE supports creating custom agents that leverage MCP capabilities, enabling users to design dedicated workflows on the fly.
228 |
229 | Theia AI and Theia IDE's MCP integration provide users with flexibility, making them powerful platforms for exploring and adapting MCP.
230 |
231 | **Learn more:**
232 |
233 | - [Theia IDE and Theia AI MCP Announcement](https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/)
234 | - [Download the AI-powered Theia IDE](https://theia-ide.org/)
235 |
236 | ### Windsurf Editor
237 |
238 | [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.
239 |
240 | **Key features:**
241 |
242 | - Revolutionary AI Flow paradigm for human-AI collaboration
243 | - Intelligent code generation and understanding
244 | - Rich development tools with multi-model support
245 |
246 | ### Zed
247 |
248 | [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.
249 |
250 | **Key features:**
251 |
252 | - Prompt templates surface as slash commands in the editor
253 | - Tool integration for enhanced coding workflows
254 | - Tight integration with editor features and workspace context
255 | - Does not support MCP resources
256 |
257 | ### OpenSumi
258 |
259 | [OpenSumi](https://github.com/opensumi/core) is a framework helps you quickly build AI Native IDE products.
260 |
261 | **Key features:**
262 |
263 | - Supports MCP tools in OpenSumi
264 | - Supports built-in IDE MCP servers and custom MCP servers
265 |
266 | ## Adding MCP support to your application
267 |
268 | 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.
269 |
270 | Benefits of adding MCP support:
271 |
272 | - Enable users to bring their own context and tools
273 | - Join a growing ecosystem of interoperable AI applications
274 | - Provide users with flexible integration options
275 | - Support local-first AI workflows
276 |
277 | 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)
278 |
279 | ## Updates and corrections
280 |
281 | 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).
282 |
283 | # Contributing
284 |
285 | Source: https://modelcontextprotocol.io/development/contributing
286 |
287 | How to participate in Model Context Protocol development
288 |
289 | 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.
290 |
291 | All contributors must adhere to our [Code of Conduct](https://github.com/modelcontextprotocol/.github/blob/main/CODE_OF_CONDUCT.md).
292 |
293 | For questions and discussions, please use [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions).
294 |
295 | # Roadmap
296 |
297 | Source: https://modelcontextprotocol.io/development/roadmap
298 |
299 | Our plans for evolving Model Context Protocol (H1 2025)
300 |
301 | 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.
302 |
303 | <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>
304 |
305 | We encourage community participation! Each section links to relevant discussions where you can learn more and contribute your thoughts.
306 |
307 | ## Remote MCP Support
308 |
309 | Our top priority is enabling [remote MCP connections](https://github.com/modelcontextprotocol/specification/discussions/102), allowing clients to securely connect to MCP servers over the internet. Key initiatives include:
310 |
311 | - [**Authentication & Authorization**](https://github.com/modelcontextprotocol/specification/discussions/64): Adding standardized auth capabilities, particularly focused on OAuth 2.0 support.
312 |
313 | - [**Service Discovery**](https://github.com/modelcontextprotocol/specification/discussions/69): Defining how clients can discover and connect to remote MCP servers.
314 |
315 | - [**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.
316 |
317 | ## Reference Implementations
318 |
319 | To help developers build with MCP, we want to offer documentation for:
320 |
321 | - **Client Examples**: Comprehensive reference client implementation(s), demonstrating all protocol features
322 | - **Protocol Drafting**: Streamlined process for proposing and incorporating new protocol features
323 |
324 | ## Distribution & Discovery
325 |
326 | Looking ahead, we're exploring ways to make MCP servers more accessible. Some areas we may investigate include:
327 |
328 | - **Package Management**: Standardized packaging format for MCP servers
329 | - **Installation Tools**: Simplified server installation across MCP clients
330 | - **Sandboxing**: Improved security through server isolation
331 | - **Server Registry**: A common directory for discovering available MCP servers
332 |
333 | ## Agent Support
334 |
335 | We're expanding MCP's capabilities for [complex agentic workflows](https://github.com/modelcontextprotocol/specification/discussions/111), particularly focusing on:
336 |
337 | - [**Hierarchical Agent Systems**](https://github.com/modelcontextprotocol/specification/discussions/94): Improved support for trees of agents through namespacing and topology awareness.
338 |
339 | - [**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.
340 |
341 | - [**Streaming Results**](https://github.com/modelcontextprotocol/specification/issues/117): Real-time updates from long-running agent operations.
342 |
343 | ## Broader Ecosystem
344 |
345 | We're also invested in:
346 |
347 | - **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.
348 | - [**Additional Modalities**](https://github.com/modelcontextprotocol/specification/discussions/88): Expanding beyond text to support audio, video, and other formats.
349 | - \[**Standardization**] Considering standardization through a standardization body.
350 |
351 | ## Get Involved
352 |
353 | 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.
354 |
355 | # What's New
356 |
357 | Source: https://modelcontextprotocol.io/development/updates
358 |
359 | The latest updates and improvements to MCP
360 |
361 | <Update label="2025-02-14" description="Java SDK released">
362 | * We're excited to announce that the Java SDK developed by Spring AI at VMware Tanzu is now
363 | the official [Java SDK](https://github.com/modelcontextprotocol/java-sdk) for MCP.
364 | This joins our existing Kotlin SDK in our growing list of supported languages.
365 | The Spring AI team will maintain the SDK as an integral part of the Model Context Protocol
366 | organization. We're thrilled to welcome them to the MCP community!
367 | </Update>
368 |
369 | <Update label="2025-01-27" description="Python SDK 1.2.1">
370 | * Version [1.2.1](https://github.com/modelcontextprotocol/python-sdk/releases/tag/v1.2.1) of the MCP Python SDK has been released,
371 | delivering important stability improvements and bug fixes.
372 | </Update>
373 |
374 | <Update label="2025-01-18" description="SDK and Server Improvements">
375 | * Simplified, express-like API in the [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)
376 | * Added 8 new clients to the [clients page](https://modelcontextprotocol.io/clients)
377 | </Update>
378 |
379 | <Update label="2025-01-03" description="SDK and Server Improvements">
380 | * FastMCP API in the [Python SDK](https://github.com/modelcontextprotocol/python-sdk)
381 | * Dockerized MCP servers in the [servers repo](https://github.com/modelcontextprotocol/servers)
382 | </Update>
383 |
384 | <Update label="2024-12-21" description="Kotlin SDK released">
385 | * Jetbrains released a Kotlin SDK for MCP!
386 | * For a sample MCP Kotlin server, check out [this repository](https://github.com/modelcontextprotocol/kotlin-sdk/tree/main/samples/kotlin-mcp-server)
387 | </Update>
388 |
389 | # Core architecture
390 |
391 | Source: https://modelcontextprotocol.io/docs/concepts/architecture
392 |
393 | Understand how MCP connects clients, servers, and LLMs
394 |
395 | 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.
396 |
397 | ## Overview
398 |
399 | MCP follows a client-server architecture where:
400 |
401 | - **Hosts** are LLM applications (like Claude Desktop or IDEs) that initiate connections
402 | - **Clients** maintain 1:1 connections with servers, inside the host application
403 | - **Servers** provide context, tools, and prompts to clients
404 |
405 | ```mermaid
406 | flowchart LR
407 | subgraph " Host "
408 | client1[MCP Client]
409 | client2[MCP Client]
410 | end
411 | subgraph "Server Process"
412 | server1[MCP Server]
413 | end
414 | subgraph "Server Process"
415 | server2[MCP Server]
416 | end
417 |
418 | client1 <-->|Transport Layer| server1
419 | client2 <-->|Transport Layer| server2
420 | ```
421 |
422 | ## Core components
423 |
424 | ### Protocol layer
425 |
426 | The protocol layer handles message framing, request/response linking, and high-level communication patterns.
427 |
428 | <Tabs>
429 | <Tab title="TypeScript">
430 | ```typescript
431 | class Protocol<Request, Notification, Result> {
432 | // Handle incoming requests
433 | setRequestHandler<T>(schema: T, handler: (request: T, extra: RequestHandlerExtra) => Promise<Result>): void
434 |
435 | // Handle incoming notifications
436 | setNotificationHandler<T>(schema: T, handler: (notification: T) => Promise<void>): void
437 |
438 | // Send requests and await responses
439 | request<T>(request: Request, schema: T, options?: RequestOptions): Promise<T>
440 |
441 | // Send one-way notifications
442 | notification(notification: Notification): Promise<void>
443 | }
444 | ```
445 |
446 | </Tab>
447 |
448 | <Tab title="Python">
449 | ```python
450 | class Session(BaseSession[RequestT, NotificationT, ResultT]):
451 | async def send_request(
452 | self,
453 | request: RequestT,
454 | result_type: type[Result]
455 | ) -> Result:
456 | """
457 | Send request and wait for response. Raises McpError if response contains error.
458 | """
459 | # Request handling implementation
460 |
461 | async def send_notification(
462 | self,
463 | notification: NotificationT
464 | ) -> None:
465 | """Send one-way notification that doesn't expect response."""
466 | # Notification handling implementation
467 |
468 | async def _received_request(
469 | self,
470 | responder: RequestResponder[ReceiveRequestT, ResultT]
471 | ) -> None:
472 | """Handle incoming request from other side."""
473 | # Request handling implementation
474 |
475 | async def _received_notification(
476 | self,
477 | notification: ReceiveNotificationT
478 | ) -> None:
479 | """Handle incoming notification from other side."""
480 | # Notification handling implementation
481 | ```
482 |
483 | </Tab>
484 | </Tabs>
485 |
486 | Key classes include:
487 |
488 | - `Protocol`
489 | - `Client`
490 | - `Server`
491 |
492 | ### Transport layer
493 |
494 | The transport layer handles the actual communication between clients and servers. MCP supports multiple transport mechanisms:
495 |
496 | 1. **Stdio transport**
497 |
498 | - Uses standard input/output for communication
499 | - Ideal for local processes
500 |
501 | 2. **HTTP with SSE transport**
502 | - Uses Server-Sent Events for server-to-client messages
503 | - HTTP POST for client-to-server messages
504 |
505 | 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.
506 |
507 | ### Message types
508 |
509 | MCP has these main types of messages:
510 |
511 | 1. **Requests** expect a response from the other side:
512 |
513 | ```typescript
514 | interface Request {
515 | method: string;
516 | params?: { ... };
517 | }
518 | ```
519 |
520 | 2. **Results** are successful responses to requests:
521 |
522 | ```typescript
523 | interface Result {
524 | [key: string]: unknown;
525 | }
526 | ```
527 |
528 | 3. **Errors** indicate that a request failed:
529 |
530 | ```typescript
531 | interface Error {
532 | code: number;
533 | message: string;
534 | data?: unknown;
535 | }
536 | ```
537 |
538 | 4. **Notifications** are one-way messages that don't expect a response:
539 | ```typescript
540 | interface Notification {
541 | method: string;
542 | params?: { ... };
543 | }
544 | ```
545 |
546 | ## Connection lifecycle
547 |
548 | ### 1. Initialization
549 |
550 | ```mermaid
551 | sequenceDiagram
552 | participant Client
553 | participant Server
554 |
555 | Client->>Server: initialize request
556 | Server->>Client: initialize response
557 | Client->>Server: initialized notification
558 |
559 | Note over Client,Server: Connection ready for use
560 | ```
561 |
562 | 1. Client sends `initialize` request with protocol version and capabilities
563 | 2. Server responds with its protocol version and capabilities
564 | 3. Client sends `initialized` notification as acknowledgment
565 | 4. Normal message exchange begins
566 |
567 | ### 2. Message exchange
568 |
569 | After initialization, the following patterns are supported:
570 |
571 | - **Request-Response**: Client or server sends requests, the other responds
572 | - **Notifications**: Either party sends one-way messages
573 |
574 | ### 3. Termination
575 |
576 | Either party can terminate the connection:
577 |
578 | - Clean shutdown via `close()`
579 | - Transport disconnection
580 | - Error conditions
581 |
582 | ## Error handling
583 |
584 | MCP defines these standard error codes:
585 |
586 | ```typescript
587 | enum ErrorCode {
588 | // Standard JSON-RPC error codes
589 | ParseError = -32700,
590 | InvalidRequest = -32600,
591 | MethodNotFound = -32601,
592 | InvalidParams = -32602,
593 | InternalError = -32603,
594 | }
595 | ```
596 |
597 | SDKs and applications can define their own error codes above -32000.
598 |
599 | Errors are propagated through:
600 |
601 | - Error responses to requests
602 | - Error events on transports
603 | - Protocol-level error handlers
604 |
605 | ## Implementation example
606 |
607 | Here's a basic example of implementing an MCP server:
608 |
609 | <Tabs>
610 | <Tab title="TypeScript">
611 | ```typescript
612 | import { Server } from "@modelcontextprotocol/sdk/server/index.js";
613 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
614 |
615 | const server = new Server({
616 | name: "example-server",
617 | version: "1.0.0"
618 | }, {
619 | capabilities: {
620 | resources: {}
621 | }
622 | });
623 |
624 | // Handle requests
625 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
626 | return {
627 | resources: [
628 | {
629 | uri: "example://resource",
630 | name: "Example Resource"
631 | }
632 | ]
633 | };
634 | });
635 |
636 | // Connect transport
637 | const transport = new StdioServerTransport();
638 | await server.connect(transport);
639 | ```
640 |
641 | </Tab>
642 |
643 | <Tab title="Python">
644 | ```python
645 | import asyncio
646 | import mcp.types as types
647 | from mcp.server import Server
648 | from mcp.server.stdio import stdio_server
649 |
650 | app = Server("example-server")
651 |
652 | @app.list_resources()
653 | async def list_resources() -> list[types.Resource]:
654 | return [
655 | types.Resource(
656 | uri="example://resource",
657 | name="Example Resource"
658 | )
659 | ]
660 |
661 | async def main():
662 | async with stdio_server() as streams:
663 | await app.run(
664 | streams[0],
665 | streams[1],
666 | app.create_initialization_options()
667 | )
668 |
669 | if __name__ == "__main__":
670 | asyncio.run(main)
671 | ```
672 |
673 | </Tab>
674 | </Tabs>
675 |
676 | ## Best practices
677 |
678 | ### Transport selection
679 |
680 | 1. **Local communication**
681 |
682 | - Use stdio transport for local processes
683 | - Efficient for same-machine communication
684 | - Simple process management
685 |
686 | 2. **Remote communication**
687 | - Use SSE for scenarios requiring HTTP compatibility
688 | - Consider security implications including authentication and authorization
689 |
690 | ### Message handling
691 |
692 | 1. **Request processing**
693 |
694 | - Validate inputs thoroughly
695 | - Use type-safe schemas
696 | - Handle errors gracefully
697 | - Implement timeouts
698 |
699 | 2. **Progress reporting**
700 |
701 | - Use progress tokens for long operations
702 | - Report progress incrementally
703 | - Include total progress when known
704 |
705 | 3. **Error management**
706 | - Use appropriate error codes
707 | - Include helpful error messages
708 | - Clean up resources on errors
709 |
710 | ## Security considerations
711 |
712 | 1. **Transport security**
713 |
714 | - Use TLS for remote connections
715 | - Validate connection origins
716 | - Implement authentication when needed
717 |
718 | 2. **Message validation**
719 |
720 | - Validate all incoming messages
721 | - Sanitize inputs
722 | - Check message size limits
723 | - Verify JSON-RPC format
724 |
725 | 3. **Resource protection**
726 |
727 | - Implement access controls
728 | - Validate resource paths
729 | - Monitor resource usage
730 | - Rate limit requests
731 |
732 | 4. **Error handling**
733 | - Don't leak sensitive information
734 | - Log security-relevant errors
735 | - Implement proper cleanup
736 | - Handle DoS scenarios
737 |
738 | ## Debugging and monitoring
739 |
740 | 1. **Logging**
741 |
742 | - Log protocol events
743 | - Track message flow
744 | - Monitor performance
745 | - Record errors
746 |
747 | 2. **Diagnostics**
748 |
749 | - Implement health checks
750 | - Monitor connection state
751 | - Track resource usage
752 | - Profile performance
753 |
754 | 3. **Testing**
755 | - Test different transports
756 | - Verify error handling
757 | - Check edge cases
758 | - Load test servers
759 |
760 | # Prompts
761 |
762 | Source: https://modelcontextprotocol.io/docs/concepts/prompts
763 |
764 | Create reusable prompt templates and workflows
765 |
766 | 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.
767 |
768 | <Note>
769 | 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.
770 | </Note>
771 |
772 | ## Overview
773 |
774 | Prompts in MCP are predefined templates that can:
775 |
776 | - Accept dynamic arguments
777 | - Include context from resources
778 | - Chain multiple interactions
779 | - Guide specific workflows
780 | - Surface as UI elements (like slash commands)
781 |
782 | ## Prompt structure
783 |
784 | Each prompt is defined with:
785 |
786 | ```typescript
787 | {
788 | name: string; // Unique identifier for the prompt
789 | description?: string; // Human-readable description
790 | arguments?: [ // Optional list of arguments
791 | {
792 | name: string; // Argument identifier
793 | description?: string; // Argument description
794 | required?: boolean; // Whether argument is required
795 | }
796 | ]
797 | }
798 | ```
799 |
800 | ## Discovering prompts
801 |
802 | Clients can discover available prompts through the `prompts/list` endpoint:
803 |
804 | ```typescript
805 | // Request
806 | {
807 | method: 'prompts/list';
808 | }
809 |
810 | // Response
811 | {
812 | prompts: [
813 | {
814 | name: 'analyze-code',
815 | description: 'Analyze code for potential improvements',
816 | arguments: [
817 | {
818 | name: 'language',
819 | description: 'Programming language',
820 | required: true,
821 | },
822 | ],
823 | },
824 | ];
825 | }
826 | ```
827 |
828 | ## Using prompts
829 |
830 | To use a prompt, clients make a `prompts/get` request:
831 |
832 | ````typescript
833 | // Request
834 | {
835 | method: "prompts/get",
836 | params: {
837 | name: "analyze-code",
838 | arguments: {
839 | language: "python"
840 | }
841 | }
842 | }
843 |
844 | // Response
845 | {
846 | description: "Analyze Python code for potential improvements",
847 | messages: [
848 | {
849 | role: "user",
850 | content: {
851 | type: "text",
852 | 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```"
853 | }
854 | }
855 | ]
856 | }
857 | ````
858 |
859 | ## Dynamic prompts
860 |
861 | Prompts can be dynamic and include:
862 |
863 | ### Embedded resource context
864 |
865 | ```json
866 | {
867 | "name": "analyze-project",
868 | "description": "Analyze project logs and code",
869 | "arguments": [
870 | {
871 | "name": "timeframe",
872 | "description": "Time period to analyze logs",
873 | "required": true
874 | },
875 | {
876 | "name": "fileUri",
877 | "description": "URI of code file to review",
878 | "required": true
879 | }
880 | ]
881 | }
882 | ```
883 |
884 | When handling the `prompts/get` request:
885 |
886 | ```json
887 | {
888 | "messages": [
889 | {
890 | "role": "user",
891 | "content": {
892 | "type": "text",
893 | "text": "Analyze these system logs and the code file for any issues:"
894 | }
895 | },
896 | {
897 | "role": "user",
898 | "content": {
899 | "type": "resource",
900 | "resource": {
901 | "uri": "logs://recent?timeframe=1h",
902 | "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",
903 | "mimeType": "text/plain"
904 | }
905 | }
906 | },
907 | {
908 | "role": "user",
909 | "content": {
910 | "type": "resource",
911 | "resource": {
912 | "uri": "file:///path/to/code.py",
913 | "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",
914 | "mimeType": "text/x-python"
915 | }
916 | }
917 | }
918 | ]
919 | }
920 | ```
921 |
922 | ### Multi-step workflows
923 |
924 | ```typescript
925 | const debugWorkflow = {
926 | name: 'debug-error',
927 | async getMessages(error: string) {
928 | return [
929 | {
930 | role: 'user',
931 | content: {
932 | type: 'text',
933 | text: `Here's an error I'm seeing: ${error}`,
934 | },
935 | },
936 | {
937 | role: 'assistant',
938 | content: {
939 | type: 'text',
940 | text: "I'll help analyze this error. What have you tried so far?",
941 | },
942 | },
943 | {
944 | role: 'user',
945 | content: {
946 | type: 'text',
947 | text: "I've tried restarting the service, but the error persists.",
948 | },
949 | },
950 | ];
951 | },
952 | };
953 | ```
954 |
955 | ## Example implementation
956 |
957 | Here's a complete example of implementing prompts in an MCP server:
958 |
959 | <Tabs>
960 | <Tab title="TypeScript">
961 | ```typescript
962 | import { Server } from "@modelcontextprotocol/sdk/server";
963 | import {
964 | ListPromptsRequestSchema,
965 | GetPromptRequestSchema
966 | } from "@modelcontextprotocol/sdk/types";
967 |
968 | const PROMPTS = {
969 | "git-commit": {
970 | name: "git-commit",
971 | description: "Generate a Git commit message",
972 | arguments: [
973 | {
974 | name: "changes",
975 | description: "Git diff or description of changes",
976 | required: true
977 | }
978 | ]
979 | },
980 | "explain-code": {
981 | name: "explain-code",
982 | description: "Explain how code works",
983 | arguments: [
984 | {
985 | name: "code",
986 | description: "Code to explain",
987 | required: true
988 | },
989 | {
990 | name: "language",
991 | description: "Programming language",
992 | required: false
993 | }
994 | ]
995 | }
996 | };
997 |
998 | const server = new Server({
999 | name: "example-prompts-server",
1000 | version: "1.0.0"
1001 | }, {
1002 | capabilities: {
1003 | prompts: {}
1004 | }
1005 | });
1006 |
1007 | // List available prompts
1008 | server.setRequestHandler(ListPromptsRequestSchema, async () => {
1009 | return {
1010 | prompts: Object.values(PROMPTS)
1011 | };
1012 | });
1013 |
1014 | // Get specific prompt
1015 | server.setRequestHandler(GetPromptRequestSchema, async (request) => {
1016 | const prompt = PROMPTS[request.params.name];
1017 | if (!prompt) {
1018 | throw new Error(`Prompt not found: ${request.params.name}`);
1019 | }
1020 |
1021 | if (request.params.name === "git-commit") {
1022 | return {
1023 | messages: [
1024 | {
1025 | role: "user",
1026 | content: {
1027 | type: "text",
1028 | text: `Generate a concise but descriptive commit message for these changes:\n\n${request.params.arguments?.changes}`
1029 | }
1030 | }
1031 | ]
1032 | };
1033 | }
1034 |
1035 | if (request.params.name === "explain-code") {
1036 | const language = request.params.arguments?.language || "Unknown";
1037 | return {
1038 | messages: [
1039 | {
1040 | role: "user",
1041 | content: {
1042 | type: "text",
1043 | text: `Explain how this ${language} code works:\n\n${request.params.arguments?.code}`
1044 | }
1045 | }
1046 | ]
1047 | };
1048 | }
1049 |
1050 | throw new Error("Prompt implementation not found");
1051 | });
1052 | ```
1053 |
1054 | </Tab>
1055 |
1056 | <Tab title="Python">
1057 | ```python
1058 | from mcp.server import Server
1059 | import mcp.types as types
1060 |
1061 | # Define available prompts
1062 | PROMPTS = {
1063 | "git-commit": types.Prompt(
1064 | name="git-commit",
1065 | description="Generate a Git commit message",
1066 | arguments=[
1067 | types.PromptArgument(
1068 | name="changes",
1069 | description="Git diff or description of changes",
1070 | required=True
1071 | )
1072 | ],
1073 | ),
1074 | "explain-code": types.Prompt(
1075 | name="explain-code",
1076 | description="Explain how code works",
1077 | arguments=[
1078 | types.PromptArgument(
1079 | name="code",
1080 | description="Code to explain",
1081 | required=True
1082 | ),
1083 | types.PromptArgument(
1084 | name="language",
1085 | description="Programming language",
1086 | required=False
1087 | )
1088 | ],
1089 | )
1090 | }
1091 |
1092 | # Initialize server
1093 | app = Server("example-prompts-server")
1094 |
1095 | @app.list_prompts()
1096 | async def list_prompts() -> list[types.Prompt]:
1097 | return list(PROMPTS.values())
1098 |
1099 | @app.get_prompt()
1100 | async def get_prompt(
1101 | name: str, arguments: dict[str, str] | None = None
1102 | ) -> types.GetPromptResult:
1103 | if name not in PROMPTS:
1104 | raise ValueError(f"Prompt not found: {name}")
1105 |
1106 | if name == "git-commit":
1107 | changes = arguments.get("changes") if arguments else ""
1108 | return types.GetPromptResult(
1109 | messages=[
1110 | types.PromptMessage(
1111 | role="user",
1112 | content=types.TextContent(
1113 | type="text",
1114 | text=f"Generate a concise but descriptive commit message "
1115 | f"for these changes:\n\n{changes}"
1116 | )
1117 | )
1118 | ]
1119 | )
1120 |
1121 | if name == "explain-code":
1122 | code = arguments.get("code") if arguments else ""
1123 | language = arguments.get("language", "Unknown") if arguments else "Unknown"
1124 | return types.GetPromptResult(
1125 | messages=[
1126 | types.PromptMessage(
1127 | role="user",
1128 | content=types.TextContent(
1129 | type="text",
1130 | text=f"Explain how this {language} code works:\n\n{code}"
1131 | )
1132 | )
1133 | ]
1134 | )
1135 |
1136 | raise ValueError("Prompt implementation not found")
1137 | ```
1138 |
1139 | </Tab>
1140 | </Tabs>
1141 |
1142 | ## Best practices
1143 |
1144 | When implementing prompts:
1145 |
1146 | 1. Use clear, descriptive prompt names
1147 | 2. Provide detailed descriptions for prompts and arguments
1148 | 3. Validate all required arguments
1149 | 4. Handle missing arguments gracefully
1150 | 5. Consider versioning for prompt templates
1151 | 6. Cache dynamic content when appropriate
1152 | 7. Implement error handling
1153 | 8. Document expected argument formats
1154 | 9. Consider prompt composability
1155 | 10. Test prompts with various inputs
1156 |
1157 | ## UI integration
1158 |
1159 | Prompts can be surfaced in client UIs as:
1160 |
1161 | - Slash commands
1162 | - Quick actions
1163 | - Context menu items
1164 | - Command palette entries
1165 | - Guided workflows
1166 | - Interactive forms
1167 |
1168 | ## Updates and changes
1169 |
1170 | Servers can notify clients about prompt changes:
1171 |
1172 | 1. Server capability: `prompts.listChanged`
1173 | 2. Notification: `notifications/prompts/list_changed`
1174 | 3. Client re-fetches prompt list
1175 |
1176 | ## Security considerations
1177 |
1178 | When implementing prompts:
1179 |
1180 | - Validate all arguments
1181 | - Sanitize user input
1182 | - Consider rate limiting
1183 | - Implement access controls
1184 | - Audit prompt usage
1185 | - Handle sensitive data appropriately
1186 | - Validate generated content
1187 | - Implement timeouts
1188 | - Consider prompt injection risks
1189 | - Document security requirements
1190 |
1191 | # Resources
1192 |
1193 | Source: https://modelcontextprotocol.io/docs/concepts/resources
1194 |
1195 | Expose data and content from your servers to LLMs
1196 |
1197 | 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.
1198 |
1199 | <Note>
1200 | Resources are designed to be **application-controlled**, meaning that the client application can decide how and when they should be used.
1201 | Different MCP clients may handle resources differently. For example:
1202 |
1203 | - Claude Desktop currently requires users to explicitly select resources before they can be used
1204 | - Other clients might automatically select resources based on heuristics
1205 | - Some implementations may even allow the AI model itself to determine which resources to use
1206 |
1207 | 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).
1208 | </Note>
1209 |
1210 | ## Overview
1211 |
1212 | Resources represent any kind of data that an MCP server wants to make available to clients. This can include:
1213 |
1214 | - File contents
1215 | - Database records
1216 | - API responses
1217 | - Live system data
1218 | - Screenshots and images
1219 | - Log files
1220 | - And more
1221 |
1222 | Each resource is identified by a unique URI and can contain either text or binary data.
1223 |
1224 | ## Resource URIs
1225 |
1226 | Resources are identified using URIs that follow this format:
1227 |
1228 | ```
1229 | [protocol]://[host]/[path]
1230 | ```
1231 |
1232 | For example:
1233 |
1234 | - `file:///home/user/documents/report.pdf`
1235 | - `postgres://database/customers/schema`
1236 | - `screen://localhost/display1`
1237 |
1238 | The protocol and path structure is defined by the MCP server implementation. Servers can define their own custom URI schemes.
1239 |
1240 | ## Resource types
1241 |
1242 | Resources can contain two types of content:
1243 |
1244 | ### Text resources
1245 |
1246 | Text resources contain UTF-8 encoded text data. These are suitable for:
1247 |
1248 | - Source code
1249 | - Configuration files
1250 | - Log files
1251 | - JSON/XML data
1252 | - Plain text
1253 |
1254 | ### Binary resources
1255 |
1256 | Binary resources contain raw binary data encoded in base64. These are suitable for:
1257 |
1258 | - Images
1259 | - PDFs
1260 | - Audio files
1261 | - Video files
1262 | - Other non-text formats
1263 |
1264 | ## Resource discovery
1265 |
1266 | Clients can discover available resources through two main methods:
1267 |
1268 | ### Direct resources
1269 |
1270 | Servers expose a list of concrete resources via the `resources/list` endpoint. Each resource includes:
1271 |
1272 | ```typescript
1273 | {
1274 | uri: string; // Unique identifier for the resource
1275 | name: string; // Human-readable name
1276 | description?: string; // Optional description
1277 | mimeType?: string; // Optional MIME type
1278 | }
1279 | ```
1280 |
1281 | ### Resource templates
1282 |
1283 | For dynamic resources, servers can expose [URI templates](https://datatracker.ietf.org/doc/html/rfc6570) that clients can use to construct valid resource URIs:
1284 |
1285 | ```typescript
1286 | {
1287 | uriTemplate: string; // URI template following RFC 6570
1288 | name: string; // Human-readable name for this type
1289 | description?: string; // Optional description
1290 | mimeType?: string; // Optional MIME type for all matching resources
1291 | }
1292 | ```
1293 |
1294 | ## Reading resources
1295 |
1296 | To read a resource, clients make a `resources/read` request with the resource URI.
1297 |
1298 | The server responds with a list of resource contents:
1299 |
1300 | ```typescript
1301 | {
1302 | contents: [
1303 | {
1304 | uri: string; // The URI of the resource
1305 | mimeType?: string; // Optional MIME type
1306 |
1307 | // One of:
1308 | text?: string; // For text resources
1309 | blob?: string; // For binary resources (base64 encoded)
1310 | }
1311 | ]
1312 | }
1313 | ```
1314 |
1315 | <Tip>
1316 | 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.
1317 | </Tip>
1318 |
1319 | ## Resource updates
1320 |
1321 | MCP supports real-time updates for resources through two mechanisms:
1322 |
1323 | ### List changes
1324 |
1325 | Servers can notify clients when their list of available resources changes via the `notifications/resources/list_changed` notification.
1326 |
1327 | ### Content changes
1328 |
1329 | Clients can subscribe to updates for specific resources:
1330 |
1331 | 1. Client sends `resources/subscribe` with resource URI
1332 | 2. Server sends `notifications/resources/updated` when the resource changes
1333 | 3. Client can fetch latest content with `resources/read`
1334 | 4. Client can unsubscribe with `resources/unsubscribe`
1335 |
1336 | ## Example implementation
1337 |
1338 | Here's a simple example of implementing resource support in an MCP server:
1339 |
1340 | <Tabs>
1341 | <Tab title="TypeScript">
1342 | ```typescript
1343 | const server = new Server({
1344 | name: "example-server",
1345 | version: "1.0.0"
1346 | }, {
1347 | capabilities: {
1348 | resources: {}
1349 | }
1350 | });
1351 |
1352 | // List available resources
1353 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
1354 | return {
1355 | resources: [
1356 | {
1357 | uri: "file:///logs/app.log",
1358 | name: "Application Logs",
1359 | mimeType: "text/plain"
1360 | }
1361 | ]
1362 | };
1363 | });
1364 |
1365 | // Read resource contents
1366 | server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
1367 | const uri = request.params.uri;
1368 |
1369 | if (uri === "file:///logs/app.log") {
1370 | const logContents = await readLogFile();
1371 | return {
1372 | contents: [
1373 | {
1374 | uri,
1375 | mimeType: "text/plain",
1376 | text: logContents
1377 | }
1378 | ]
1379 | };
1380 | }
1381 |
1382 | throw new Error("Resource not found");
1383 | });
1384 | ```
1385 |
1386 | </Tab>
1387 |
1388 | <Tab title="Python">
1389 | ```python
1390 | app = Server("example-server")
1391 |
1392 | @app.list_resources()
1393 | async def list_resources() -> list[types.Resource]:
1394 | return [
1395 | types.Resource(
1396 | uri="file:///logs/app.log",
1397 | name="Application Logs",
1398 | mimeType="text/plain"
1399 | )
1400 | ]
1401 |
1402 | @app.read_resource()
1403 | async def read_resource(uri: AnyUrl) -> str:
1404 | if str(uri) == "file:///logs/app.log":
1405 | log_contents = await read_log_file()
1406 | return log_contents
1407 |
1408 | raise ValueError("Resource not found")
1409 |
1410 | # Start server
1411 | async with stdio_server() as streams:
1412 | await app.run(
1413 | streams[0],
1414 | streams[1],
1415 | app.create_initialization_options()
1416 | )
1417 | ```
1418 |
1419 | </Tab>
1420 | </Tabs>
1421 |
1422 | ## Best practices
1423 |
1424 | When implementing resource support:
1425 |
1426 | 1. Use clear, descriptive resource names and URIs
1427 | 2. Include helpful descriptions to guide LLM understanding
1428 | 3. Set appropriate MIME types when known
1429 | 4. Implement resource templates for dynamic content
1430 | 5. Use subscriptions for frequently changing resources
1431 | 6. Handle errors gracefully with clear error messages
1432 | 7. Consider pagination for large resource lists
1433 | 8. Cache resource contents when appropriate
1434 | 9. Validate URIs before processing
1435 | 10. Document your custom URI schemes
1436 |
1437 | ## Security considerations
1438 |
1439 | When exposing resources:
1440 |
1441 | - Validate all resource URIs
1442 | - Implement appropriate access controls
1443 | - Sanitize file paths to prevent directory traversal
1444 | - Be cautious with binary data handling
1445 | - Consider rate limiting for resource reads
1446 | - Audit resource access
1447 | - Encrypt sensitive data in transit
1448 | - Validate MIME types
1449 | - Implement timeouts for long-running reads
1450 | - Handle resource cleanup appropriately
1451 |
1452 | # Roots
1453 |
1454 | Source: https://modelcontextprotocol.io/docs/concepts/roots
1455 |
1456 | Understanding roots in MCP
1457 |
1458 | 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.
1459 |
1460 | ## What are Roots?
1461 |
1462 | 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.
1463 |
1464 | For example, roots could be:
1465 |
1466 | ```
1467 | file:///home/user/projects/myapp
1468 | https://api.example.com/v1
1469 | ```
1470 |
1471 | ## Why Use Roots?
1472 |
1473 | Roots serve several important purposes:
1474 |
1475 | 1. **Guidance**: They inform servers about relevant resources and locations
1476 | 2. **Clarity**: Roots make it clear which resources are part of your workspace
1477 | 3. **Organization**: Multiple roots let you work with different resources simultaneously
1478 |
1479 | ## How Roots Work
1480 |
1481 | When a client supports roots, it:
1482 |
1483 | 1. Declares the `roots` capability during connection
1484 | 2. Provides a list of suggested roots to the server
1485 | 3. Notifies the server when roots change (if supported)
1486 |
1487 | While roots are informational and not strictly enforcing, servers should:
1488 |
1489 | 1. Respect the provided roots
1490 | 2. Use root URIs to locate and access resources
1491 | 3. Prioritize operations within root boundaries
1492 |
1493 | ## Common Use Cases
1494 |
1495 | Roots are commonly used to define:
1496 |
1497 | - Project directories
1498 | - Repository locations
1499 | - API endpoints
1500 | - Configuration locations
1501 | - Resource boundaries
1502 |
1503 | ## Best Practices
1504 |
1505 | When working with roots:
1506 |
1507 | 1. Only suggest necessary resources
1508 | 2. Use clear, descriptive names for roots
1509 | 3. Monitor root accessibility
1510 | 4. Handle root changes gracefully
1511 |
1512 | ## Example
1513 |
1514 | Here's how a typical MCP client might expose roots:
1515 |
1516 | ```json
1517 | {
1518 | "roots": [
1519 | {
1520 | "uri": "file:///home/user/projects/frontend",
1521 | "name": "Frontend Repository"
1522 | },
1523 | {
1524 | "uri": "https://api.example.com/v1",
1525 | "name": "API Endpoint"
1526 | }
1527 | ]
1528 | }
1529 | ```
1530 |
1531 | This configuration suggests the server focus on both a local repository and an API endpoint while keeping them logically separated.
1532 |
1533 | # Sampling
1534 |
1535 | Source: https://modelcontextprotocol.io/docs/concepts/sampling
1536 |
1537 | Let your servers request completions from LLMs
1538 |
1539 | 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.
1540 |
1541 | <Info>
1542 | This feature of MCP is not yet supported in the Claude Desktop client.
1543 | </Info>
1544 |
1545 | ## How sampling works
1546 |
1547 | The sampling flow follows these steps:
1548 |
1549 | 1. Server sends a `sampling/createMessage` request to the client
1550 | 2. Client reviews the request and can modify it
1551 | 3. Client samples from an LLM
1552 | 4. Client reviews the completion
1553 | 5. Client returns the result to the server
1554 |
1555 | This human-in-the-loop design ensures users maintain control over what the LLM sees and generates.
1556 |
1557 | ## Message format
1558 |
1559 | Sampling requests use a standardized message format:
1560 |
1561 | ```typescript
1562 | {
1563 | messages: [
1564 | {
1565 | role: "user" | "assistant",
1566 | content: {
1567 | type: "text" | "image",
1568 |
1569 | // For text:
1570 | text?: string,
1571 |
1572 | // For images:
1573 | data?: string, // base64 encoded
1574 | mimeType?: string
1575 | }
1576 | }
1577 | ],
1578 | modelPreferences?: {
1579 | hints?: [{
1580 | name?: string // Suggested model name/family
1581 | }],
1582 | costPriority?: number, // 0-1, importance of minimizing cost
1583 | speedPriority?: number, // 0-1, importance of low latency
1584 | intelligencePriority?: number // 0-1, importance of capabilities
1585 | },
1586 | systemPrompt?: string,
1587 | includeContext?: "none" | "thisServer" | "allServers",
1588 | temperature?: number,
1589 | maxTokens: number,
1590 | stopSequences?: string[],
1591 | metadata?: Record<string, unknown>
1592 | }
1593 | ```
1594 |
1595 | ## Request parameters
1596 |
1597 | ### Messages
1598 |
1599 | The `messages` array contains the conversation history to send to the LLM. Each message has:
1600 |
1601 | - `role`: Either "user" or "assistant"
1602 | - `content`: The message content, which can be:
1603 | - Text content with a `text` field
1604 | - Image content with `data` (base64) and `mimeType` fields
1605 |
1606 | ### Model preferences
1607 |
1608 | The `modelPreferences` object allows servers to specify their model selection preferences:
1609 |
1610 | - `hints`: Array of model name suggestions that clients can use to select an appropriate model:
1611 |
1612 | - `name`: String that can match full or partial model names (e.g. "claude-3", "sonnet")
1613 | - Clients may map hints to equivalent models from different providers
1614 | - Multiple hints are evaluated in preference order
1615 |
1616 | - Priority values (0-1 normalized):
1617 | - `costPriority`: Importance of minimizing costs
1618 | - `speedPriority`: Importance of low latency response
1619 | - `intelligencePriority`: Importance of advanced model capabilities
1620 |
1621 | Clients make the final model selection based on these preferences and their available models.
1622 |
1623 | ### System prompt
1624 |
1625 | An optional `systemPrompt` field allows servers to request a specific system prompt. The client may modify or ignore this.
1626 |
1627 | ### Context inclusion
1628 |
1629 | The `includeContext` parameter specifies what MCP context to include:
1630 |
1631 | - `"none"`: No additional context
1632 | - `"thisServer"`: Include context from the requesting server
1633 | - `"allServers"`: Include context from all connected MCP servers
1634 |
1635 | The client controls what context is actually included.
1636 |
1637 | ### Sampling parameters
1638 |
1639 | Fine-tune the LLM sampling with:
1640 |
1641 | - `temperature`: Controls randomness (0.0 to 1.0)
1642 | - `maxTokens`: Maximum tokens to generate
1643 | - `stopSequences`: Array of sequences that stop generation
1644 | - `metadata`: Additional provider-specific parameters
1645 |
1646 | ## Response format
1647 |
1648 | The client returns a completion result:
1649 |
1650 | ```typescript
1651 | {
1652 | model: string, // Name of the model used
1653 | stopReason?: "endTurn" | "stopSequence" | "maxTokens" | string,
1654 | role: "user" | "assistant",
1655 | content: {
1656 | type: "text" | "image",
1657 | text?: string,
1658 | data?: string,
1659 | mimeType?: string
1660 | }
1661 | }
1662 | ```
1663 |
1664 | ## Example request
1665 |
1666 | Here's an example of requesting sampling from a client:
1667 |
1668 | ```json
1669 | {
1670 | "method": "sampling/createMessage",
1671 | "params": {
1672 | "messages": [
1673 | {
1674 | "role": "user",
1675 | "content": {
1676 | "type": "text",
1677 | "text": "What files are in the current directory?"
1678 | }
1679 | }
1680 | ],
1681 | "systemPrompt": "You are a helpful file system assistant.",
1682 | "includeContext": "thisServer",
1683 | "maxTokens": 100
1684 | }
1685 | }
1686 | ```
1687 |
1688 | ## Best practices
1689 |
1690 | When implementing sampling:
1691 |
1692 | 1. Always provide clear, well-structured prompts
1693 | 2. Handle both text and image content appropriately
1694 | 3. Set reasonable token limits
1695 | 4. Include relevant context through `includeContext`
1696 | 5. Validate responses before using them
1697 | 6. Handle errors gracefully
1698 | 7. Consider rate limiting sampling requests
1699 | 8. Document expected sampling behavior
1700 | 9. Test with various model parameters
1701 | 10. Monitor sampling costs
1702 |
1703 | ## Human in the loop controls
1704 |
1705 | Sampling is designed with human oversight in mind:
1706 |
1707 | ### For prompts
1708 |
1709 | - Clients should show users the proposed prompt
1710 | - Users should be able to modify or reject prompts
1711 | - System prompts can be filtered or modified
1712 | - Context inclusion is controlled by the client
1713 |
1714 | ### For completions
1715 |
1716 | - Clients should show users the completion
1717 | - Users should be able to modify or reject completions
1718 | - Clients can filter or modify completions
1719 | - Users control which model is used
1720 |
1721 | ## Security considerations
1722 |
1723 | When implementing sampling:
1724 |
1725 | - Validate all message content
1726 | - Sanitize sensitive information
1727 | - Implement appropriate rate limits
1728 | - Monitor sampling usage
1729 | - Encrypt data in transit
1730 | - Handle user data privacy
1731 | - Audit sampling requests
1732 | - Control cost exposure
1733 | - Implement timeouts
1734 | - Handle model errors gracefully
1735 |
1736 | ## Common patterns
1737 |
1738 | ### Agentic workflows
1739 |
1740 | Sampling enables agentic patterns like:
1741 |
1742 | - Reading and analyzing resources
1743 | - Making decisions based on context
1744 | - Generating structured data
1745 | - Handling multi-step tasks
1746 | - Providing interactive assistance
1747 |
1748 | ### Context management
1749 |
1750 | Best practices for context:
1751 |
1752 | - Request minimal necessary context
1753 | - Structure context clearly
1754 | - Handle context size limits
1755 | - Update context as needed
1756 | - Clean up stale context
1757 |
1758 | ### Error handling
1759 |
1760 | Robust error handling should:
1761 |
1762 | - Catch sampling failures
1763 | - Handle timeout errors
1764 | - Manage rate limits
1765 | - Validate responses
1766 | - Provide fallback behaviors
1767 | - Log errors appropriately
1768 |
1769 | ## Limitations
1770 |
1771 | Be aware of these limitations:
1772 |
1773 | - Sampling depends on client capabilities
1774 | - Users control sampling behavior
1775 | - Context size has limits
1776 | - Rate limits may apply
1777 | - Costs should be considered
1778 | - Model availability varies
1779 | - Response times vary
1780 | - Not all content types supported
1781 |
1782 | # Tools
1783 |
1784 | Source: https://modelcontextprotocol.io/docs/concepts/tools
1785 |
1786 | Enable LLMs to perform actions through your server
1787 |
1788 | 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.
1789 |
1790 | <Note>
1791 | 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).
1792 | </Note>
1793 |
1794 | ## Overview
1795 |
1796 | 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:
1797 |
1798 | - **Discovery**: Clients can list available tools through the `tools/list` endpoint
1799 | - **Invocation**: Tools are called using the `tools/call` endpoint, where servers perform the requested operation and return results
1800 | - **Flexibility**: Tools can range from simple calculations to complex API interactions
1801 |
1802 | 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.
1803 |
1804 | ## Tool definition structure
1805 |
1806 | Each tool is defined with the following structure:
1807 |
1808 | ```typescript
1809 | {
1810 | name: string; // Unique identifier for the tool
1811 | description?: string; // Human-readable description
1812 | inputSchema: { // JSON Schema for the tool's parameters
1813 | type: "object",
1814 | properties: { ... } // Tool-specific parameters
1815 | }
1816 | }
1817 | ```
1818 |
1819 | ## Implementing tools
1820 |
1821 | Here's an example of implementing a basic tool in an MCP server:
1822 |
1823 | <Tabs>
1824 | <Tab title="TypeScript">
1825 | ```typescript
1826 | const server = new Server({
1827 | name: "example-server",
1828 | version: "1.0.0"
1829 | }, {
1830 | capabilities: {
1831 | tools: {}
1832 | }
1833 | });
1834 |
1835 | // Define available tools
1836 | server.setRequestHandler(ListToolsRequestSchema, async () => {
1837 | return {
1838 | tools: [{
1839 | name: "calculate_sum",
1840 | description: "Add two numbers together",
1841 | inputSchema: {
1842 | type: "object",
1843 | properties: {
1844 | a: { type: "number" },
1845 | b: { type: "number" }
1846 | },
1847 | required: ["a", "b"]
1848 | }
1849 | }]
1850 | };
1851 | });
1852 |
1853 | // Handle tool execution
1854 | server.setRequestHandler(CallToolRequestSchema, async (request) => {
1855 | if (request.params.name === "calculate_sum") {
1856 | const { a, b } = request.params.arguments;
1857 | return {
1858 | content: [
1859 | {
1860 | type: "text",
1861 | text: String(a + b)
1862 | }
1863 | ]
1864 | };
1865 | }
1866 | throw new Error("Tool not found");
1867 | });
1868 | ```
1869 |
1870 | </Tab>
1871 |
1872 | <Tab title="Python">
1873 | ```python
1874 | app = Server("example-server")
1875 |
1876 | @app.list_tools()
1877 | async def list_tools() -> list[types.Tool]:
1878 | return [
1879 | types.Tool(
1880 | name="calculate_sum",
1881 | description="Add two numbers together",
1882 | inputSchema={
1883 | "type": "object",
1884 | "properties": {
1885 | "a": {"type": "number"},
1886 | "b": {"type": "number"}
1887 | },
1888 | "required": ["a", "b"]
1889 | }
1890 | )
1891 | ]
1892 |
1893 | @app.call_tool()
1894 | async def call_tool(
1895 | name: str,
1896 | arguments: dict
1897 | ) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
1898 | if name == "calculate_sum":
1899 | a = arguments["a"]
1900 | b = arguments["b"]
1901 | result = a + b
1902 | return [types.TextContent(type="text", text=str(result))]
1903 | raise ValueError(f"Tool not found: {name}")
1904 | ```
1905 |
1906 | </Tab>
1907 | </Tabs>
1908 |
1909 | ## Example tool patterns
1910 |
1911 | Here are some examples of types of tools that a server could provide:
1912 |
1913 | ### System operations
1914 |
1915 | Tools that interact with the local system:
1916 |
1917 | ```typescript
1918 | {
1919 | name: "execute_command",
1920 | description: "Run a shell command",
1921 | inputSchema: {
1922 | type: "object",
1923 | properties: {
1924 | command: { type: "string" },
1925 | args: { type: "array", items: { type: "string" } }
1926 | }
1927 | }
1928 | }
1929 | ```
1930 |
1931 | ### API integrations
1932 |
1933 | Tools that wrap external APIs:
1934 |
1935 | ```typescript
1936 | {
1937 | name: "github_create_issue",
1938 | description: "Create a GitHub issue",
1939 | inputSchema: {
1940 | type: "object",
1941 | properties: {
1942 | title: { type: "string" },
1943 | body: { type: "string" },
1944 | labels: { type: "array", items: { type: "string" } }
1945 | }
1946 | }
1947 | }
1948 | ```
1949 |
1950 | ### Data processing
1951 |
1952 | Tools that transform or analyze data:
1953 |
1954 | ```typescript
1955 | {
1956 | name: "analyze_csv",
1957 | description: "Analyze a CSV file",
1958 | inputSchema: {
1959 | type: "object",
1960 | properties: {
1961 | filepath: { type: "string" },
1962 | operations: {
1963 | type: "array",
1964 | items: {
1965 | enum: ["sum", "average", "count"]
1966 | }
1967 | }
1968 | }
1969 | }
1970 | }
1971 | ```
1972 |
1973 | ## Best practices
1974 |
1975 | When implementing tools:
1976 |
1977 | 1. Provide clear, descriptive names and descriptions
1978 | 2. Use detailed JSON Schema definitions for parameters
1979 | 3. Include examples in tool descriptions to demonstrate how the model should use them
1980 | 4. Implement proper error handling and validation
1981 | 5. Use progress reporting for long operations
1982 | 6. Keep tool operations focused and atomic
1983 | 7. Document expected return value structures
1984 | 8. Implement proper timeouts
1985 | 9. Consider rate limiting for resource-intensive operations
1986 | 10. Log tool usage for debugging and monitoring
1987 |
1988 | ## Security considerations
1989 |
1990 | When exposing tools:
1991 |
1992 | ### Input validation
1993 |
1994 | - Validate all parameters against the schema
1995 | - Sanitize file paths and system commands
1996 | - Validate URLs and external identifiers
1997 | - Check parameter sizes and ranges
1998 | - Prevent command injection
1999 |
2000 | ### Access control
2001 |
2002 | - Implement authentication where needed
2003 | - Use appropriate authorization checks
2004 | - Audit tool usage
2005 | - Rate limit requests
2006 | - Monitor for abuse
2007 |
2008 | ### Error handling
2009 |
2010 | - Don't expose internal errors to clients
2011 | - Log security-relevant errors
2012 | - Handle timeouts appropriately
2013 | - Clean up resources after errors
2014 | - Validate return values
2015 |
2016 | ## Tool discovery and updates
2017 |
2018 | MCP supports dynamic tool discovery:
2019 |
2020 | 1. Clients can list available tools at any time
2021 | 2. Servers can notify clients when tools change using `notifications/tools/list_changed`
2022 | 3. Tools can be added or removed during runtime
2023 | 4. Tool definitions can be updated (though this should be done carefully)
2024 |
2025 | ## Error handling
2026 |
2027 | 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:
2028 |
2029 | 1. Set `isError` to `true` in the result
2030 | 2. Include error details in the `content` array
2031 |
2032 | Here's an example of proper error handling for tools:
2033 |
2034 | <Tabs>
2035 | <Tab title="TypeScript">
2036 | ```typescript
2037 | try {
2038 | // Tool operation
2039 | const result = performOperation();
2040 | return {
2041 | content: [
2042 | {
2043 | type: "text",
2044 | text: `Operation successful: ${result}`
2045 | }
2046 | ]
2047 | };
2048 | } catch (error) {
2049 | return {
2050 | isError: true,
2051 | content: [
2052 | {
2053 | type: "text",
2054 | text: `Error: ${error.message}`
2055 | }
2056 | ]
2057 | };
2058 | }
2059 | ```
2060 | </Tab>
2061 |
2062 | <Tab title="Python">
2063 | ```python
2064 | try:
2065 | # Tool operation
2066 | result = perform_operation()
2067 | return types.CallToolResult(
2068 | content=[
2069 | types.TextContent(
2070 | type="text",
2071 | text=f"Operation successful: {result}"
2072 | )
2073 | ]
2074 | )
2075 | except Exception as error:
2076 | return types.CallToolResult(
2077 | isError=True,
2078 | content=[
2079 | types.TextContent(
2080 | type="text",
2081 | text=f"Error: {str(error)}"
2082 | )
2083 | ]
2084 | )
2085 | ```
2086 | </Tab>
2087 | </Tabs>
2088 |
2089 | This approach allows the LLM to see that an error occurred and potentially take corrective action or request human intervention.
2090 |
2091 | ## Testing tools
2092 |
2093 | A comprehensive testing strategy for MCP tools should cover:
2094 |
2095 | - **Functional testing**: Verify tools execute correctly with valid inputs and handle invalid inputs appropriately
2096 | - **Integration testing**: Test tool interaction with external systems using both real and mocked dependencies
2097 | - **Security testing**: Validate authentication, authorization, input sanitization, and rate limiting
2098 | - **Performance testing**: Check behavior under load, timeout handling, and resource cleanup
2099 | - **Error handling**: Ensure tools properly report errors through the MCP protocol and clean up resources
2100 |
2101 | # Transports
2102 |
2103 | Source: https://modelcontextprotocol.io/docs/concepts/transports
2104 |
2105 | Learn about MCP's communication mechanisms
2106 |
2107 | 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.
2108 |
2109 | ## Message Format
2110 |
2111 | 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.
2112 |
2113 | There are three types of JSON-RPC messages used:
2114 |
2115 | ### Requests
2116 |
2117 | ```typescript
2118 | {
2119 | jsonrpc: "2.0",
2120 | id: number | string,
2121 | method: string,
2122 | params?: object
2123 | }
2124 | ```
2125 |
2126 | ### Responses
2127 |
2128 | ```typescript
2129 | {
2130 | jsonrpc: "2.0",
2131 | id: number | string,
2132 | result?: object,
2133 | error?: {
2134 | code: number,
2135 | message: string,
2136 | data?: unknown
2137 | }
2138 | }
2139 | ```
2140 |
2141 | ### Notifications
2142 |
2143 | ```typescript
2144 | {
2145 | jsonrpc: "2.0",
2146 | method: string,
2147 | params?: object
2148 | }
2149 | ```
2150 |
2151 | ## Built-in Transport Types
2152 |
2153 | MCP includes two standard transport implementations:
2154 |
2155 | ### Standard Input/Output (stdio)
2156 |
2157 | The stdio transport enables communication through standard input and output streams. This is particularly useful for local integrations and command-line tools.
2158 |
2159 | Use stdio when:
2160 |
2161 | - Building command-line tools
2162 | - Implementing local integrations
2163 | - Needing simple process communication
2164 | - Working with shell scripts
2165 |
2166 | <Tabs>
2167 | <Tab title="TypeScript (Server)">
2168 | ```typescript
2169 | const server = new Server({
2170 | name: "example-server",
2171 | version: "1.0.0"
2172 | }, {
2173 | capabilities: {}
2174 | });
2175 |
2176 | const transport = new StdioServerTransport();
2177 | await server.connect(transport);
2178 | ```
2179 |
2180 | </Tab>
2181 |
2182 | <Tab title="TypeScript (Client)">
2183 | ```typescript
2184 | const client = new Client({
2185 | name: "example-client",
2186 | version: "1.0.0"
2187 | }, {
2188 | capabilities: {}
2189 | });
2190 |
2191 | const transport = new StdioClientTransport({
2192 | command: "./server",
2193 | args: ["--option", "value"]
2194 | });
2195 | await client.connect(transport);
2196 | ```
2197 |
2198 | </Tab>
2199 |
2200 | <Tab title="Python (Server)">
2201 | ```python
2202 | app = Server("example-server")
2203 |
2204 | async with stdio_server() as streams:
2205 | await app.run(
2206 | streams[0],
2207 | streams[1],
2208 | app.create_initialization_options()
2209 | )
2210 | ```
2211 |
2212 | </Tab>
2213 |
2214 | <Tab title="Python (Client)">
2215 | ```python
2216 | params = StdioServerParameters(
2217 | command="./server",
2218 | args=["--option", "value"]
2219 | )
2220 |
2221 | async with stdio_client(params) as streams:
2222 | async with ClientSession(streams[0], streams[1]) as session:
2223 | await session.initialize()
2224 | ```
2225 |
2226 | </Tab>
2227 | </Tabs>
2228 |
2229 | ### Server-Sent Events (SSE)
2230 |
2231 | SSE transport enables server-to-client streaming with HTTP POST requests for client-to-server communication.
2232 |
2233 | Use SSE when:
2234 |
2235 | - Only server-to-client streaming is needed
2236 | - Working with restricted networks
2237 | - Implementing simple updates
2238 |
2239 | <Tabs>
2240 | <Tab title="TypeScript (Server)">
2241 | ```typescript
2242 | import express from "express";
2243 |
2244 | const app = express();
2245 |
2246 | const server = new Server({
2247 | name: "example-server",
2248 | version: "1.0.0"
2249 | }, {
2250 | capabilities: {}
2251 | });
2252 |
2253 | let transport: SSEServerTransport | null = null;
2254 |
2255 | app.get("/sse", (req, res) => {
2256 | transport = new SSEServerTransport("/messages", res);
2257 | server.connect(transport);
2258 | });
2259 |
2260 | app.post("/messages", (req, res) => {
2261 | if (transport) {
2262 | transport.handlePostMessage(req, res);
2263 | }
2264 | });
2265 |
2266 | app.listen(3000);
2267 | ```
2268 |
2269 | </Tab>
2270 |
2271 | <Tab title="TypeScript (Client)">
2272 | ```typescript
2273 | const client = new Client({
2274 | name: "example-client",
2275 | version: "1.0.0"
2276 | }, {
2277 | capabilities: {}
2278 | });
2279 |
2280 | const transport = new SSEClientTransport(
2281 | new URL("http://localhost:3000/sse")
2282 | );
2283 | await client.connect(transport);
2284 | ```
2285 |
2286 | </Tab>
2287 |
2288 | <Tab title="Python (Server)">
2289 | ```python
2290 | from mcp.server.sse import SseServerTransport
2291 | from starlette.applications import Starlette
2292 | from starlette.routing import Route
2293 |
2294 | app = Server("example-server")
2295 | sse = SseServerTransport("/messages")
2296 |
2297 | async def handle_sse(scope, receive, send):
2298 | async with sse.connect_sse(scope, receive, send) as streams:
2299 | await app.run(streams[0], streams[1], app.create_initialization_options())
2300 |
2301 | async def handle_messages(scope, receive, send):
2302 | await sse.handle_post_message(scope, receive, send)
2303 |
2304 | starlette_app = Starlette(
2305 | routes=[
2306 | Route("/sse", endpoint=handle_sse),
2307 | Route("/messages", endpoint=handle_messages, methods=["POST"]),
2308 | ]
2309 | )
2310 | ```
2311 |
2312 | </Tab>
2313 |
2314 | <Tab title="Python (Client)">
2315 | ```python
2316 | async with sse_client("http://localhost:8000/sse") as streams:
2317 | async with ClientSession(streams[0], streams[1]) as session:
2318 | await session.initialize()
2319 | ```
2320 | </Tab>
2321 | </Tabs>
2322 |
2323 | ## Custom Transports
2324 |
2325 | MCP makes it easy to implement custom transports for specific needs. Any transport implementation just needs to conform to the Transport interface:
2326 |
2327 | You can implement custom transports for:
2328 |
2329 | - Custom network protocols
2330 | - Specialized communication channels
2331 | - Integration with existing systems
2332 | - Performance optimization
2333 |
2334 | <Tabs>
2335 | <Tab title="TypeScript">
2336 | ```typescript
2337 | interface Transport {
2338 | // Start processing messages
2339 | start(): Promise<void>;
2340 |
2341 | // Send a JSON-RPC message
2342 | send(message: JSONRPCMessage): Promise<void>;
2343 |
2344 | // Close the connection
2345 | close(): Promise<void>;
2346 |
2347 | // Callbacks
2348 | onclose?: () => void;
2349 | onerror?: (error: Error) => void;
2350 | onmessage?: (message: JSONRPCMessage) => void;
2351 | }
2352 | ```
2353 |
2354 | </Tab>
2355 |
2356 | <Tab title="Python">
2357 | Note that while MCP Servers are often implemented with asyncio, we recommend
2358 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2359 |
2360 | ```python
2361 | @contextmanager
2362 | async def create_transport(
2363 | read_stream: MemoryObjectReceiveStream[JSONRPCMessage | Exception],
2364 | write_stream: MemoryObjectSendStream[JSONRPCMessage]
2365 | ):
2366 | """
2367 | Transport interface for MCP.
2368 |
2369 | Args:
2370 | read_stream: Stream to read incoming messages from
2371 | write_stream: Stream to write outgoing messages to
2372 | """
2373 | async with anyio.create_task_group() as tg:
2374 | try:
2375 | # Start processing messages
2376 | tg.start_soon(lambda: process_messages(read_stream))
2377 |
2378 | # Send messages
2379 | async with write_stream:
2380 | yield write_stream
2381 |
2382 | except Exception as exc:
2383 | # Handle errors
2384 | raise exc
2385 | finally:
2386 | # Clean up
2387 | tg.cancel_scope.cancel()
2388 | await write_stream.aclose()
2389 | await read_stream.aclose()
2390 | ```
2391 |
2392 | </Tab>
2393 | </Tabs>
2394 |
2395 | ## Error Handling
2396 |
2397 | Transport implementations should handle various error scenarios:
2398 |
2399 | 1. Connection errors
2400 | 2. Message parsing errors
2401 | 3. Protocol errors
2402 | 4. Network timeouts
2403 | 5. Resource cleanup
2404 |
2405 | Example error handling:
2406 |
2407 | <Tabs>
2408 | <Tab title="TypeScript">
2409 | ```typescript
2410 | class ExampleTransport implements Transport {
2411 | async start() {
2412 | try {
2413 | // Connection logic
2414 | } catch (error) {
2415 | this.onerror?.(new Error(`Failed to connect: ${error}`));
2416 | throw error;
2417 | }
2418 | }
2419 |
2420 | async send(message: JSONRPCMessage) {
2421 | try {
2422 | // Sending logic
2423 | } catch (error) {
2424 | this.onerror?.(new Error(`Failed to send message: ${error}`));
2425 | throw error;
2426 | }
2427 | }
2428 | }
2429 | ```
2430 |
2431 | </Tab>
2432 |
2433 | <Tab title="Python">
2434 | Note that while MCP Servers are often implemented with asyncio, we recommend
2435 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2436 |
2437 | ```python
2438 | @contextmanager
2439 | async def example_transport(scope: Scope, receive: Receive, send: Send):
2440 | try:
2441 | # Create streams for bidirectional communication
2442 | read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
2443 | write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
2444 |
2445 | async def message_handler():
2446 | try:
2447 | async with read_stream_writer:
2448 | # Message handling logic
2449 | pass
2450 | except Exception as exc:
2451 | logger.error(f"Failed to handle message: {exc}")
2452 | raise exc
2453 |
2454 | async with anyio.create_task_group() as tg:
2455 | tg.start_soon(message_handler)
2456 | try:
2457 | # Yield streams for communication
2458 | yield read_stream, write_stream
2459 | except Exception as exc:
2460 | logger.error(f"Transport error: {exc}")
2461 | raise exc
2462 | finally:
2463 | tg.cancel_scope.cancel()
2464 | await write_stream.aclose()
2465 | await read_stream.aclose()
2466 | except Exception as exc:
2467 | logger.error(f"Failed to initialize transport: {exc}")
2468 | raise exc
2469 | ```
2470 |
2471 | </Tab>
2472 | </Tabs>
2473 |
2474 | ## Best Practices
2475 |
2476 | When implementing or using MCP transport:
2477 |
2478 | 1. Handle connection lifecycle properly
2479 | 2. Implement proper error handling
2480 | 3. Clean up resources on connection close
2481 | 4. Use appropriate timeouts
2482 | 5. Validate messages before sending
2483 | 6. Log transport events for debugging
2484 | 7. Implement reconnection logic when appropriate
2485 | 8. Handle backpressure in message queues
2486 | 9. Monitor connection health
2487 | 10. Implement proper security measures
2488 |
2489 | ## Security Considerations
2490 |
2491 | When implementing transport:
2492 |
2493 | ### Authentication and Authorization
2494 |
2495 | - Implement proper authentication mechanisms
2496 | - Validate client credentials
2497 | - Use secure token handling
2498 | - Implement authorization checks
2499 |
2500 | ### Data Security
2501 |
2502 | - Use TLS for network transport
2503 | - Encrypt sensitive data
2504 | - Validate message integrity
2505 | - Implement message size limits
2506 | - Sanitize input data
2507 |
2508 | ### Network Security
2509 |
2510 | - Implement rate limiting
2511 | - Use appropriate timeouts
2512 | - Handle denial of service scenarios
2513 | - Monitor for unusual patterns
2514 | - Implement proper firewall rules
2515 |
2516 | ## Debugging Transport
2517 |
2518 | Tips for debugging transport issues:
2519 |
2520 | 1. Enable debug logging
2521 | 2. Monitor message flow
2522 | 3. Check connection states
2523 | 4. Validate message formats
2524 | 5. Test error scenarios
2525 | 6. Use network analysis tools
2526 | 7. Implement health checks
2527 | 8. Monitor resource usage
2528 | 9. Test edge cases
2529 | 10. Use proper error tracking
2530 |
2531 | # Debugging
2532 |
2533 | Source: https://modelcontextprotocol.io/docs/tools/debugging
2534 |
2535 | A comprehensive guide to debugging Model Context Protocol (MCP) integrations
2536 |
2537 | 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.
2538 |
2539 | <Info>
2540 | This guide is for macOS. Guides for other platforms are coming soon.
2541 | </Info>
2542 |
2543 | ## Debugging tools overview
2544 |
2545 | MCP provides several tools for debugging at different levels:
2546 |
2547 | 1. **MCP Inspector**
2548 |
2549 | - Interactive debugging interface
2550 | - Direct server testing
2551 | - See the [Inspector guide](/docs/tools/inspector) for details
2552 |
2553 | 2. **Claude Desktop Developer Tools**
2554 |
2555 | - Integration testing
2556 | - Log collection
2557 | - Chrome DevTools integration
2558 |
2559 | 3. **Server Logging**
2560 | - Custom logging implementations
2561 | - Error tracking
2562 | - Performance monitoring
2563 |
2564 | ## Debugging in Claude Desktop
2565 |
2566 | ### Checking server status
2567 |
2568 | The Claude.app interface provides basic server status information:
2569 |
2570 | 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:
2571 |
2572 | - Connected servers
2573 | - Available prompts and resources
2574 |
2575 | 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:
2576 | - Tools made available to the model
2577 |
2578 | ### Viewing logs
2579 |
2580 | Review detailed MCP logs from Claude Desktop:
2581 |
2582 | ```bash
2583 | # Follow logs in real-time
2584 | tail -n 20 -F ~/Library/Logs/Claude/mcp*.log
2585 | ```
2586 |
2587 | The logs capture:
2588 |
2589 | - Server connection events
2590 | - Configuration issues
2591 | - Runtime errors
2592 | - Message exchanges
2593 |
2594 | ### Using Chrome DevTools
2595 |
2596 | Access Chrome's developer tools inside Claude Desktop to investigate client-side errors:
2597 |
2598 | 1. Create a `developer_settings.json` file with `allowDevTools` set to true:
2599 |
2600 | ```bash
2601 | echo '{"allowDevTools": true}' > ~/Library/Application\ Support/Claude/developer_settings.json
2602 | ```
2603 |
2604 | 2. Open DevTools: `Command-Option-Shift-i`
2605 |
2606 | Note: You'll see two DevTools windows:
2607 |
2608 | - Main content window
2609 | - App title bar window
2610 |
2611 | Use the Console panel to inspect client-side errors.
2612 |
2613 | Use the Network panel to inspect:
2614 |
2615 | - Message payloads
2616 | - Connection timing
2617 |
2618 | ## Common issues
2619 |
2620 | ### Working directory
2621 |
2622 | When using MCP servers with Claude Desktop:
2623 |
2624 | - 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
2625 | - Always use absolute paths in your configuration and `.env` files to ensure reliable operation
2626 | - For testing servers directly via command line, the working directory will be where you run the command
2627 |
2628 | For example in `claude_desktop_config.json`, use:
2629 |
2630 | ```json
2631 | {
2632 | "command": "npx",
2633 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/username/data"]
2634 | }
2635 | ```
2636 |
2637 | Instead of relative paths like `./data`
2638 |
2639 | ### Environment variables
2640 |
2641 | MCP servers inherit only a subset of environment variables automatically, like `USER`, `HOME`, and `PATH`.
2642 |
2643 | To override the default variables or provide your own, you can specify an `env` key in `claude_desktop_config.json`:
2644 |
2645 | ```json
2646 | {
2647 | "myserver": {
2648 | "command": "mcp-server-myapp",
2649 | "env": {
2650 | "MYAPP_API_KEY": "some_key"
2651 | }
2652 | }
2653 | }
2654 | ```
2655 |
2656 | ### Server initialization
2657 |
2658 | Common initialization problems:
2659 |
2660 | 1. **Path Issues**
2661 |
2662 | - Incorrect server executable path
2663 | - Missing required files
2664 | - Permission problems
2665 | - Try using an absolute path for `command`
2666 |
2667 | 2. **Configuration Errors**
2668 |
2669 | - Invalid JSON syntax
2670 | - Missing required fields
2671 | - Type mismatches
2672 |
2673 | 3. **Environment Problems**
2674 | - Missing environment variables
2675 | - Incorrect variable values
2676 | - Permission restrictions
2677 |
2678 | ### Connection problems
2679 |
2680 | When servers fail to connect:
2681 |
2682 | 1. Check Claude Desktop logs
2683 | 2. Verify server process is running
2684 | 3. Test standalone with [Inspector](/docs/tools/inspector)
2685 | 4. Verify protocol compatibility
2686 |
2687 | ## Implementing logging
2688 |
2689 | ### Server-side logging
2690 |
2691 | 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.
2692 |
2693 | <Warning>
2694 | Local MCP servers should not log messages to stdout (standard out), as this will interfere with protocol operation.
2695 | </Warning>
2696 |
2697 | For all [transports](/docs/concepts/transports), you can also provide logging to the client by sending a log message notification:
2698 |
2699 | <Tabs>
2700 | <Tab title="Python">
2701 | ```python
2702 | server.request_context.session.send_log_message(
2703 | level="info",
2704 | data="Server started successfully",
2705 | )
2706 | ```
2707 | </Tab>
2708 |
2709 | <Tab title="TypeScript">
2710 | ```typescript
2711 | server.sendLoggingMessage({
2712 | level: "info",
2713 | data: "Server started successfully",
2714 | });
2715 | ```
2716 | </Tab>
2717 | </Tabs>
2718 |
2719 | Important events to log:
2720 |
2721 | - Initialization steps
2722 | - Resource access
2723 | - Tool execution
2724 | - Error conditions
2725 | - Performance metrics
2726 |
2727 | ### Client-side logging
2728 |
2729 | In client applications:
2730 |
2731 | 1. Enable debug logging
2732 | 2. Monitor network traffic
2733 | 3. Track message exchanges
2734 | 4. Record error states
2735 |
2736 | ## Debugging workflow
2737 |
2738 | ### Development cycle
2739 |
2740 | 1. Initial Development
2741 |
2742 | - Use [Inspector](/docs/tools/inspector) for basic testing
2743 | - Implement core functionality
2744 | - Add logging points
2745 |
2746 | 2. Integration Testing
2747 | - Test in Claude Desktop
2748 | - Monitor logs
2749 | - Check error handling
2750 |
2751 | ### Testing changes
2752 |
2753 | To test changes efficiently:
2754 |
2755 | - **Configuration changes**: Restart Claude Desktop
2756 | - **Server code changes**: Use Command-R to reload
2757 | - **Quick iteration**: Use [Inspector](/docs/tools/inspector) during development
2758 |
2759 | ## Best practices
2760 |
2761 | ### Logging strategy
2762 |
2763 | 1. **Structured Logging**
2764 |
2765 | - Use consistent formats
2766 | - Include context
2767 | - Add timestamps
2768 | - Track request IDs
2769 |
2770 | 2. **Error Handling**
2771 |
2772 | - Log stack traces
2773 | - Include error context
2774 | - Track error patterns
2775 | - Monitor recovery
2776 |
2777 | 3. **Performance Tracking**
2778 | - Log operation timing
2779 | - Monitor resource usage
2780 | - Track message sizes
2781 | - Measure latency
2782 |
2783 | ### Security considerations
2784 |
2785 | When debugging:
2786 |
2787 | 1. **Sensitive Data**
2788 |
2789 | - Sanitize logs
2790 | - Protect credentials
2791 | - Mask personal information
2792 |
2793 | 2. **Access Control**
2794 | - Verify permissions
2795 | - Check authentication
2796 | - Monitor access patterns
2797 |
2798 | ## Getting help
2799 |
2800 | When encountering issues:
2801 |
2802 | 1. **First Steps**
2803 |
2804 | - Check server logs
2805 | - Test with [Inspector](/docs/tools/inspector)
2806 | - Review configuration
2807 | - Verify environment
2808 |
2809 | 2. **Support Channels**
2810 |
2811 | - GitHub issues
2812 | - GitHub discussions
2813 |
2814 | 3. **Providing Information**
2815 | - Log excerpts
2816 | - Configuration files
2817 | - Steps to reproduce
2818 | - Environment details
2819 |
2820 | ## Next steps
2821 |
2822 | <CardGroup cols={2}>
2823 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
2824 | Learn to use the MCP Inspector
2825 | </Card>
2826 | </CardGroup>
2827 |
2828 | # Inspector
2829 |
2830 | Source: https://modelcontextprotocol.io/docs/tools/inspector
2831 |
2832 | In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers
2833 |
2834 | 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.
2835 |
2836 | ## Getting started
2837 |
2838 | ### Installation and basic usage
2839 |
2840 | The Inspector runs directly through `npx` without requiring installation:
2841 |
2842 | ```bash
2843 | npx @modelcontextprotocol/inspector <command>
2844 | ```
2845 |
2846 | ```bash
2847 | npx @modelcontextprotocol/inspector <command> <arg1> <arg2>
2848 | ```
2849 |
2850 | #### Inspecting servers from NPM or PyPi
2851 |
2852 | A common way to start server packages from [NPM](https://npmjs.com) or [PyPi](https://pypi.com).
2853 |
2854 | <Tabs>
2855 | <Tab title="NPM package">
2856 | ```bash
2857 | npx -y @modelcontextprotocol/inspector npx <package-name> <args>
2858 | # For example
2859 | npx -y @modelcontextprotocol/inspector npx server-postgres postgres://127.0.0.1/testdb
2860 | ```
2861 | </Tab>
2862 |
2863 | <Tab title="PyPi package">
2864 | ```bash
2865 | npx @modelcontextprotocol/inspector uvx <package-name> <args>
2866 | # For example
2867 | npx @modelcontextprotocol/inspector uvx mcp-server-git --repository ~/code/mcp/servers.git
2868 | ```
2869 | </Tab>
2870 | </Tabs>
2871 |
2872 | #### Inspecting locally developed servers
2873 |
2874 | To inspect servers locally developed or downloaded as a repository, the most common
2875 | way is:
2876 |
2877 | <Tabs>
2878 | <Tab title="TypeScript">
2879 | ```bash
2880 | npx @modelcontextprotocol/inspector node path/to/server/index.js args...
2881 | ```
2882 | </Tab>
2883 |
2884 | <Tab title="Python">
2885 | ```bash
2886 | npx @modelcontextprotocol/inspector \
2887 | uv \
2888 | --directory path/to/server \
2889 | run \
2890 | package-name \
2891 | args...
2892 | ```
2893 | </Tab>
2894 | </Tabs>
2895 |
2896 | Please carefully read any attached README for the most accurate instructions.
2897 |
2898 | ## Feature overview
2899 |
2900 | <Frame caption="The MCP Inspector interface">
2901 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/mcp-inspector.png" />
2902 | </Frame>
2903 |
2904 | The Inspector provides several features for interacting with your MCP server:
2905 |
2906 | ### Server connection pane
2907 |
2908 | - Allows selecting the [transport](/docs/concepts/transports) for connecting to the server
2909 | - For local servers, supports customizing the command-line arguments and environment
2910 |
2911 | ### Resources tab
2912 |
2913 | - Lists all available resources
2914 | - Shows resource metadata (MIME types, descriptions)
2915 | - Allows resource content inspection
2916 | - Supports subscription testing
2917 |
2918 | ### Prompts tab
2919 |
2920 | - Displays available prompt templates
2921 | - Shows prompt arguments and descriptions
2922 | - Enables prompt testing with custom arguments
2923 | - Previews generated messages
2924 |
2925 | ### Tools tab
2926 |
2927 | - Lists available tools
2928 | - Shows tool schemas and descriptions
2929 | - Enables tool testing with custom inputs
2930 | - Displays tool execution results
2931 |
2932 | ### Notifications pane
2933 |
2934 | - Presents all logs recorded from the server
2935 | - Shows notifications received from the server
2936 |
2937 | ## Best practices
2938 |
2939 | ### Development workflow
2940 |
2941 | 1. Start Development
2942 |
2943 | - Launch Inspector with your server
2944 | - Verify basic connectivity
2945 | - Check capability negotiation
2946 |
2947 | 2. Iterative testing
2948 |
2949 | - Make server changes
2950 | - Rebuild the server
2951 | - Reconnect the Inspector
2952 | - Test affected features
2953 | - Monitor messages
2954 |
2955 | 3. Test edge cases
2956 | - Invalid inputs
2957 | - Missing prompt arguments
2958 | - Concurrent operations
2959 | - Verify error handling and error responses
2960 |
2961 | ## Next steps
2962 |
2963 | <CardGroup cols={2}>
2964 | <Card title="Inspector Repository" icon="github" href="https://github.com/modelcontextprotocol/inspector">
2965 | Check out the MCP Inspector source code
2966 | </Card>
2967 |
2968 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
2969 | Learn about broader debugging strategies
2970 | </Card>
2971 | </CardGroup>
2972 |
2973 | # Example Servers
2974 |
2975 | Source: https://modelcontextprotocol.io/examples
2976 |
2977 | A list of example servers and implementations
2978 |
2979 | 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.
2980 |
2981 | ## Reference implementations
2982 |
2983 | These official reference servers demonstrate core MCP features and SDK usage:
2984 |
2985 | ### Data and file systems
2986 |
2987 | - **[Filesystem](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)** - Secure file operations with configurable access controls
2988 | - **[PostgreSQL](https://github.com/modelcontextprotocol/servers/tree/main/src/postgres)** - Read-only database access with schema inspection capabilities
2989 | - **[SQLite](https://github.com/modelcontextprotocol/servers/tree/main/src/sqlite)** - Database interaction and business intelligence features
2990 | - **[Google Drive](https://github.com/modelcontextprotocol/servers/tree/main/src/gdrive)** - File access and search capabilities for Google Drive
2991 |
2992 | ### Development tools
2993 |
2994 | - **[Git](https://github.com/modelcontextprotocol/servers/tree/main/src/git)** - Tools to read, search, and manipulate Git repositories
2995 | - **[GitHub](https://github.com/modelcontextprotocol/servers/tree/main/src/github)** - Repository management, file operations, and GitHub API integration
2996 | - **[GitLab](https://github.com/modelcontextprotocol/servers/tree/main/src/gitlab)** - GitLab API integration enabling project management
2997 | - **[Sentry](https://github.com/modelcontextprotocol/servers/tree/main/src/sentry)** - Retrieving and analyzing issues from Sentry.io
2998 |
2999 | ### Web and browser automation
3000 |
3001 | - **[Brave Search](https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search)** - Web and local search using Brave's Search API
3002 | - **[Fetch](https://github.com/modelcontextprotocol/servers/tree/main/src/fetch)** - Web content fetching and conversion optimized for LLM usage
3003 | - **[Puppeteer](https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer)** - Browser automation and web scraping capabilities
3004 |
3005 | ### Productivity and communication
3006 |
3007 | - **[Slack](https://github.com/modelcontextprotocol/servers/tree/main/src/slack)** - Channel management and messaging capabilities
3008 | - **[Google Maps](https://github.com/modelcontextprotocol/servers/tree/main/src/google-maps)** - Location services, directions, and place details
3009 | - **[Memory](https://github.com/modelcontextprotocol/servers/tree/main/src/memory)** - Knowledge graph-based persistent memory system
3010 |
3011 | ### AI and specialized tools
3012 |
3013 | - **[EverArt](https://github.com/modelcontextprotocol/servers/tree/main/src/everart)** - AI image generation using various models
3014 | - **[Sequential Thinking](https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking)** - Dynamic problem-solving through thought sequences
3015 | - **[AWS KB Retrieval](https://github.com/modelcontextprotocol/servers/tree/main/src/aws-kb-retrieval-server)** - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime
3016 |
3017 | ## Official integrations
3018 |
3019 | These MCP servers are maintained by companies for their platforms:
3020 |
3021 | - **[Axiom](https://github.com/axiomhq/mcp-server-axiom)** - Query and analyze logs, traces, and event data using natural language
3022 | - **[Browserbase](https://github.com/browserbase/mcp-server-browserbase)** - Automate browser interactions in the cloud
3023 | - **[Cloudflare](https://github.com/cloudflare/mcp-server-cloudflare)** - Deploy and manage resources on the Cloudflare developer platform
3024 | - **[E2B](https://github.com/e2b-dev/mcp-server)** - Execute code in secure cloud sandboxes
3025 | - **[Neon](https://github.com/neondatabase/mcp-server-neon)** - Interact with the Neon serverless Postgres platform
3026 | - **[Obsidian Markdown Notes](https://github.com/calclavia/mcp-obsidian)** - Read and search through Markdown notes in Obsidian vaults
3027 | - **[Qdrant](https://github.com/qdrant/mcp-server-qdrant/)** - Implement semantic memory using the Qdrant vector search engine
3028 | - **[Raygun](https://github.com/MindscapeHQ/mcp-server-raygun)** - Access crash reporting and monitoring data
3029 | - **[Search1API](https://github.com/fatwang2/search1api-mcp)** - Unified API for search, crawling, and sitemaps
3030 | - **[Stripe](https://github.com/stripe/agent-toolkit)** - Interact with the Stripe API
3031 | - **[Tinybird](https://github.com/tinybirdco/mcp-tinybird)** - Interface with the Tinybird serverless ClickHouse platform
3032 |
3033 | ## Community highlights
3034 |
3035 | A growing ecosystem of community-developed servers extends MCP's capabilities:
3036 |
3037 | - **[Docker](https://github.com/ckreiling/mcp-server-docker)** - Manage containers, images, volumes, and networks
3038 | - **[Kubernetes](https://github.com/Flux159/mcp-server-kubernetes)** - Manage pods, deployments, and services
3039 | - **[Linear](https://github.com/jerhadf/linear-mcp-server)** - Project management and issue tracking
3040 | - **[Snowflake](https://github.com/datawiz168/mcp-snowflake-service)** - Interact with Snowflake databases
3041 | - **[Spotify](https://github.com/varunneal/spotify-mcp)** - Control Spotify playback and manage playlists
3042 | - **[Todoist](https://github.com/abhiz123/todoist-mcp-server)** - Task management integration
3043 |
3044 | > **Note:** Community servers are untested and should be used at your own risk. They are not affiliated with or endorsed by Anthropic.
3045 |
3046 | For a complete list of community servers, visit the [MCP Servers Repository](https://github.com/modelcontextprotocol/servers).
3047 |
3048 | ## Getting started
3049 |
3050 | ### Using reference servers
3051 |
3052 | TypeScript-based servers can be used directly with `npx`:
3053 |
3054 | ```bash
3055 | npx -y @modelcontextprotocol/server-memory
3056 | ```
3057 |
3058 | Python-based servers can be used with `uvx` (recommended) or `pip`:
3059 |
3060 | ```bash
3061 | # Using uvx
3062 | uvx mcp-server-git
3063 |
3064 | # Using pip
3065 | pip install mcp-server-git
3066 | python -m mcp_server_git
3067 | ```
3068 |
3069 | ### Configuring with Claude
3070 |
3071 | To use an MCP server with Claude, add it to your configuration:
3072 |
3073 | ```json
3074 | {
3075 | "mcpServers": {
3076 | "memory": {
3077 | "command": "npx",
3078 | "args": ["-y", "@modelcontextprotocol/server-memory"]
3079 | },
3080 | "filesystem": {
3081 | "command": "npx",
3082 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
3083 | },
3084 | "github": {
3085 | "command": "npx",
3086 | "args": ["-y", "@modelcontextprotocol/server-github"],
3087 | "env": {
3088 | "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
3089 | }
3090 | }
3091 | }
3092 | }
3093 | ```
3094 |
3095 | ## Additional resources
3096 |
3097 | - [MCP Servers Repository](https://github.com/modelcontextprotocol/servers) - Complete collection of reference implementations and community servers
3098 | - [Awesome MCP Servers](https://github.com/punkpeye/awesome-mcp-servers) - Curated list of MCP servers
3099 | - [MCP CLI](https://github.com/wong2/mcp-cli) - Command-line inspector for testing MCP servers
3100 | - [MCP Get](https://mcp-get.com) - Tool for installing and managing MCP servers
3101 | - [Supergateway](https://github.com/supercorp-ai/supergateway) - Run MCP stdio servers over SSE
3102 |
3103 | Visit our [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions) to engage with the MCP community.
3104 |
3105 | # Introduction
3106 |
3107 | Source: https://modelcontextprotocol.io/introduction
3108 |
3109 | Get started with the Model Context Protocol (MCP)
3110 |
3111 | <Note>Java SDK released! Check out [what else is new.](/development/updates)</Note>
3112 |
3113 | 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.
3114 |
3115 | ## Why MCP?
3116 |
3117 | MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:
3118 |
3119 | - A growing list of pre-built integrations that your LLM can directly plug into
3120 | - The flexibility to switch between LLM providers and vendors
3121 | - Best practices for securing your data within your infrastructure
3122 |
3123 | ### General architecture
3124 |
3125 | At its core, MCP follows a client-server architecture where a host application can connect to multiple servers:
3126 |
3127 | ```mermaid
3128 | flowchart LR
3129 | subgraph "Your Computer"
3130 | Host["Host with MCP Client\n(Claude, IDEs, Tools)"]
3131 | S1["MCP Server A"]
3132 | S2["MCP Server B"]
3133 | S3["MCP Server C"]
3134 | Host <-->|"MCP Protocol"| S1
3135 | Host <-->|"MCP Protocol"| S2
3136 | Host <-->|"MCP Protocol"| S3
3137 | S1 <--> D1[("Local\nData Source A")]
3138 | S2 <--> D2[("Local\nData Source B")]
3139 | end
3140 | subgraph "Internet"
3141 | S3 <-->|"Web APIs"| D3[("Remote\nService C")]
3142 | end
3143 | ```
3144 |
3145 | - **MCP Hosts**: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
3146 | - **MCP Clients**: Protocol clients that maintain 1:1 connections with servers
3147 | - **MCP Servers**: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
3148 | - **Local Data Sources**: Your computer's files, databases, and services that MCP servers can securely access
3149 | - **Remote Services**: External systems available over the internet (e.g., through APIs) that MCP servers can connect to
3150 |
3151 | ## Get started
3152 |
3153 | Choose the path that best fits your needs:
3154 |
3155 | #### Quick Starts
3156 |
3157 | <CardGroup cols={2}>
3158 | <Card title="For Server Developers" icon="bolt" href="/quickstart/server">
3159 | Get started building your own server to use in Claude for Desktop and other clients
3160 | </Card>
3161 |
3162 | <Card title="For Client Developers" icon="bolt" href="/quickstart/client">
3163 | Get started building your own client that can integrate with all MCP servers
3164 | </Card>
3165 |
3166 | <Card title="For Claude Desktop Users" icon="bolt" href="/quickstart/user">
3167 | Get started using pre-built servers in Claude for Desktop
3168 | </Card>
3169 | </CardGroup>
3170 |
3171 | #### Examples
3172 |
3173 | <CardGroup cols={2}>
3174 | <Card title="Example Servers" icon="grid" href="/examples">
3175 | Check out our gallery of official MCP servers and implementations
3176 | </Card>
3177 |
3178 | <Card title="Example Clients" icon="cubes" href="/clients">
3179 | View the list of clients that support MCP integrations
3180 | </Card>
3181 | </CardGroup>
3182 |
3183 | ## Tutorials
3184 |
3185 | <CardGroup cols={2}>
3186 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
3187 | Learn how to use LLMs like Claude to speed up your MCP development
3188 | </Card>
3189 |
3190 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
3191 | Learn how to effectively debug MCP servers and integrations
3192 | </Card>
3193 |
3194 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
3195 | Test and inspect your MCP servers with our interactive debugging tool
3196 | </Card>
3197 | </CardGroup>
3198 |
3199 | ## Explore MCP
3200 |
3201 | Dive deeper into MCP's core concepts and capabilities:
3202 |
3203 | <CardGroup cols={2}>
3204 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3205 | Understand how MCP connects clients, servers, and LLMs
3206 | </Card>
3207 |
3208 | <Card title="Resources" icon="database" href="/docs/concepts/resources">
3209 | Expose data and content from your servers to LLMs
3210 | </Card>
3211 |
3212 | <Card title="Prompts" icon="message" href="/docs/concepts/prompts">
3213 | Create reusable prompt templates and workflows
3214 | </Card>
3215 |
3216 | <Card title="Tools" icon="wrench" href="/docs/concepts/tools">
3217 | Enable LLMs to perform actions through your server
3218 | </Card>
3219 |
3220 | <Card title="Sampling" icon="robot" href="/docs/concepts/sampling">
3221 | Let your servers request completions from LLMs
3222 | </Card>
3223 |
3224 | <Card title="Transports" icon="network-wired" href="/docs/concepts/transports">
3225 | Learn about MCP's communication mechanism
3226 | </Card>
3227 | </CardGroup>
3228 |
3229 | ## Contributing
3230 |
3231 | Want to contribute? Check out our [Contributing Guide](/development/contributing) to learn how you can help improve MCP.
3232 |
3233 | ## Support and Feedback
3234 |
3235 | Here's how to get help or provide feedback:
3236 |
3237 | - 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)
3238 | - For discussions or Q\&A about the MCP specification, use the [specification discussions](https://github.com/modelcontextprotocol/specification/discussions)
3239 | - For discussions or Q\&A about other MCP open source components, use the [organization discussions](https://github.com/orgs/modelcontextprotocol/discussions)
3240 | - For bug reports, feature requests, and questions related to Claude.app and claude.ai's MCP integration, please email [[email protected]](mailto:[email protected])
3241 |
3242 | # For Client Developers
3243 |
3244 | Source: https://modelcontextprotocol.io/quickstart/client
3245 |
3246 | Get started building your own client that can integrate with all MCP servers.
3247 |
3248 | 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.
3249 |
3250 | <Tabs>
3251 | <Tab title="Python">
3252 | [You can find the complete code for this tutorial here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/mcp-client)
3253 |
3254 | ## System Requirements
3255 |
3256 | Before starting, ensure your system meets these requirements:
3257 |
3258 | * Mac or Windows computer
3259 | * Latest Python version installed
3260 | * Latest version of `uv` installed
3261 |
3262 | ## Setting Up Your Environment
3263 |
3264 | First, create a new Python project with `uv`:
3265 |
3266 | ```bash
3267 | # Create project directory
3268 | uv init mcp-client
3269 | cd mcp-client
3270 |
3271 | # Create virtual environment
3272 | uv venv
3273 |
3274 | # Activate virtual environment
3275 | # On Windows:
3276 | .venv\Scripts\activate
3277 | # On Unix or MacOS:
3278 | source .venv/bin/activate
3279 |
3280 | # Install required packages
3281 | uv add mcp anthropic python-dotenv
3282 |
3283 | # Remove boilerplate files
3284 | rm hello.py
3285 |
3286 | # Create our main file
3287 | touch client.py
3288 | ```
3289 |
3290 | ## Setting Up Your API Key
3291 |
3292 | You'll need an Anthropic API key from the [Anthropic Console](https://console.anthropic.com/settings/keys).
3293 |
3294 | Create a `.env` file to store it:
3295 |
3296 | ```bash
3297 | # Create .env file
3298 | touch .env
3299 | ```
3300 |
3301 | Add your key to the `.env` file:
3302 |
3303 | ```bash
3304 | ANTHROPIC_API_KEY=<your key here>
3305 | ```
3306 |
3307 | Add `.env` to your `.gitignore`:
3308 |
3309 | ```bash
3310 | echo ".env" >> .gitignore
3311 | ```
3312 |
3313 | <Warning>
3314 | Make sure you keep your `ANTHROPIC_API_KEY` secure!
3315 | </Warning>
3316 |
3317 | ## Creating the Client
3318 |
3319 | ### Basic Client Structure
3320 |
3321 | First, let's set up our imports and create the basic client class:
3322 |
3323 | ```python
3324 | import asyncio
3325 | from typing import Optional
3326 | from contextlib import AsyncExitStack
3327 |
3328 | from mcp import ClientSession, StdioServerParameters
3329 | from mcp.client.stdio import stdio_client
3330 |
3331 | from anthropic import Anthropic
3332 | from dotenv import load_dotenv
3333 |
3334 | load_dotenv() # load environment variables from .env
3335 |
3336 | class MCPClient:
3337 | def __init__(self):
3338 | # Initialize session and client objects
3339 | self.session: Optional[ClientSession] = None
3340 | self.exit_stack = AsyncExitStack()
3341 | self.anthropic = Anthropic()
3342 | # methods will go here
3343 | ```
3344 |
3345 | ### Server Connection Management
3346 |
3347 | Next, we'll implement the method to connect to an MCP server:
3348 |
3349 | ```python
3350 | async def connect_to_server(self, server_script_path: str):
3351 | """Connect to an MCP server
3352 |
3353 | Args:
3354 | server_script_path: Path to the server script (.py or .js)
3355 | """
3356 | is_python = server_script_path.endswith('.py')
3357 | is_js = server_script_path.endswith('.js')
3358 | if not (is_python or is_js):
3359 | raise ValueError("Server script must be a .py or .js file")
3360 |
3361 | command = "python" if is_python else "node"
3362 | server_params = StdioServerParameters(
3363 | command=command,
3364 | args=[server_script_path],
3365 | env=None
3366 | )
3367 |
3368 | stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
3369 | self.stdio, self.write = stdio_transport
3370 | self.session = await self.exit_stack.enter_async_context(ClientSession(self.stdio, self.write))
3371 |
3372 | await self.session.initialize()
3373 |
3374 | # List available tools
3375 | response = await self.session.list_tools()
3376 | tools = response.tools
3377 | print("\nConnected to server with tools:", [tool.name for tool in tools])
3378 | ```
3379 |
3380 | ### Query Processing Logic
3381 |
3382 | Now let's add the core functionality for processing queries and handling tool calls:
3383 |
3384 | ```python
3385 | async def process_query(self, query: str) -> str:
3386 | """Process a query using Claude and available tools"""
3387 | messages = [
3388 | {
3389 | "role": "user",
3390 | "content": query
3391 | }
3392 | ]
3393 |
3394 | response = await self.session.list_tools()
3395 | available_tools = [{
3396 | "name": tool.name,
3397 | "description": tool.description,
3398 | "input_schema": tool.inputSchema
3399 | } for tool in response.tools]
3400 |
3401 | # Initial Claude API call
3402 | response = self.anthropic.messages.create(
3403 | model="claude-3-5-sonnet-20241022",
3404 | max_tokens=1000,
3405 | messages=messages,
3406 | tools=available_tools
3407 | )
3408 |
3409 | # Process response and handle tool calls
3410 | tool_results = []
3411 | final_text = []
3412 |
3413 | assistant_message_content = []
3414 | for content in response.content:
3415 | if content.type == 'text':
3416 | final_text.append(content.text)
3417 | assistant_message_content.append(content)
3418 | elif content.type == 'tool_use':
3419 | tool_name = content.name
3420 | tool_args = content.input
3421 |
3422 | # Execute tool call
3423 | result = await self.session.call_tool(tool_name, tool_args)
3424 | tool_results.append({"call": tool_name, "result": result})
3425 | final_text.append(f"[Calling tool {tool_name} with args {tool_args}]")
3426 |
3427 | assistant_message_content.append(content)
3428 | messages.append({
3429 | "role": "assistant",
3430 | "content": assistant_message_content
3431 | })
3432 | messages.append({
3433 | "role": "user",
3434 | "content": [
3435 | {
3436 | "type": "tool_result",
3437 | "tool_use_id": content.id,
3438 | "content": result.content
3439 | }
3440 | ]
3441 | })
3442 |
3443 | # Get next response from Claude
3444 | response = self.anthropic.messages.create(
3445 | model="claude-3-5-sonnet-20241022",
3446 | max_tokens=1000,
3447 | messages=messages,
3448 | tools=available_tools
3449 | )
3450 |
3451 | final_text.append(response.content[0].text)
3452 |
3453 | return "\n".join(final_text)
3454 | ```
3455 |
3456 | ### Interactive Chat Interface
3457 |
3458 | Now we'll add the chat loop and cleanup functionality:
3459 |
3460 | ```python
3461 | async def chat_loop(self):
3462 | """Run an interactive chat loop"""
3463 | print("\nMCP Client Started!")
3464 | print("Type your queries or 'quit' to exit.")
3465 |
3466 | while True:
3467 | try:
3468 | query = input("\nQuery: ").strip()
3469 |
3470 | if query.lower() == 'quit':
3471 | break
3472 |
3473 | response = await self.process_query(query)
3474 | print("\n" + response)
3475 |
3476 | except Exception as e:
3477 | print(f"\nError: {str(e)}")
3478 |
3479 | async def cleanup(self):
3480 | """Clean up resources"""
3481 | await self.exit_stack.aclose()
3482 | ```
3483 |
3484 | ### Main Entry Point
3485 |
3486 | Finally, we'll add the main execution logic:
3487 |
3488 | ```python
3489 | async def main():
3490 | if len(sys.argv) < 2:
3491 | print("Usage: python client.py <path_to_server_script>")
3492 | sys.exit(1)
3493 |
3494 | client = MCPClient()
3495 | try:
3496 | await client.connect_to_server(sys.argv[1])
3497 | await client.chat_loop()
3498 | finally:
3499 | await client.cleanup()
3500 |
3501 | if __name__ == "__main__":
3502 | import sys
3503 | asyncio.run(main())
3504 | ```
3505 |
3506 | You can find the complete `client.py` file [here.](https://gist.github.com/zckly/f3f28ea731e096e53b39b47bf0a2d4b1)
3507 |
3508 | ## Key Components Explained
3509 |
3510 | ### 1. Client Initialization
3511 |
3512 | * The `MCPClient` class initializes with session management and API clients
3513 | * Uses `AsyncExitStack` for proper resource management
3514 | * Configures the Anthropic client for Claude interactions
3515 |
3516 | ### 2. Server Connection
3517 |
3518 | * Supports both Python and Node.js servers
3519 | * Validates server script type
3520 | * Sets up proper communication channels
3521 | * Initializes the session and lists available tools
3522 |
3523 | ### 3. Query Processing
3524 |
3525 | * Maintains conversation context
3526 | * Handles Claude's responses and tool calls
3527 | * Manages the message flow between Claude and tools
3528 | * Combines results into a coherent response
3529 |
3530 | ### 4. Interactive Interface
3531 |
3532 | * Provides a simple command-line interface
3533 | * Handles user input and displays responses
3534 | * Includes basic error handling
3535 | * Allows graceful exit
3536 |
3537 | ### 5. Resource Management
3538 |
3539 | * Proper cleanup of resources
3540 | * Error handling for connection issues
3541 | * Graceful shutdown procedures
3542 |
3543 | ## Common Customization Points
3544 |
3545 | 1. **Tool Handling**
3546 | * Modify `process_query()` to handle specific tool types
3547 | * Add custom error handling for tool calls
3548 | * Implement tool-specific response formatting
3549 |
3550 | 2. **Response Processing**
3551 | * Customize how tool results are formatted
3552 | * Add response filtering or transformation
3553 | * Implement custom logging
3554 |
3555 | 3. **User Interface**
3556 | * Add a GUI or web interface
3557 | * Implement rich console output
3558 | * Add command history or auto-completion
3559 |
3560 | ## Running the Client
3561 |
3562 | To run your client with any MCP server:
3563 |
3564 | ```bash
3565 | uv run client.py path/to/server.py # python server
3566 | uv run client.py path/to/build/index.js # node server
3567 | ```
3568 |
3569 | <Note>
3570 | 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`
3571 | </Note>
3572 |
3573 | The client will:
3574 |
3575 | 1. Connect to the specified server
3576 | 2. List available tools
3577 | 3. Start an interactive chat session where you can:
3578 | * Enter queries
3579 | * See tool executions
3580 | * Get responses from Claude
3581 |
3582 | Here's an example of what it should look like if connected to the weather server from the server quickstart:
3583 |
3584 | <Frame>
3585 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/client-claude-cli-python.png" />
3586 | </Frame>
3587 |
3588 | ## How It Works
3589 |
3590 | When you submit a query:
3591 |
3592 | 1. The client gets the list of available tools from the server
3593 | 2. Your query is sent to Claude along with tool descriptions
3594 | 3. Claude decides which tools (if any) to use
3595 | 4. The client executes any requested tool calls through the server
3596 | 5. Results are sent back to Claude
3597 | 6. Claude provides a natural language response
3598 | 7. The response is displayed to you
3599 |
3600 | ## Best practices
3601 |
3602 | 1. **Error Handling**
3603 | * Always wrap tool calls in try-catch blocks
3604 | * Provide meaningful error messages
3605 | * Gracefully handle connection issues
3606 |
3607 | 2. **Resource Management**
3608 | * Use `AsyncExitStack` for proper cleanup
3609 | * Close connections when done
3610 | * Handle server disconnections
3611 |
3612 | 3. **Security**
3613 | * Store API keys securely in `.env`
3614 | * Validate server responses
3615 | * Be cautious with tool permissions
3616 |
3617 | ## Troubleshooting
3618 |
3619 | ### Server Path Issues
3620 |
3621 | * Double-check the path to your server script is correct
3622 | * Use the absolute path if the relative path isn't working
3623 | * For Windows users, make sure to use forward slashes (/) or escaped backslashes (\\) in the path
3624 | * Verify the server file has the correct extension (.py for Python or .js for Node.js)
3625 |
3626 | Example of correct path usage:
3627 |
3628 | ```bash
3629 | # Relative path
3630 | uv run client.py ./server/weather.py
3631 |
3632 | # Absolute path
3633 | uv run client.py /Users/username/projects/mcp-server/weather.py
3634 |
3635 | # Windows path (either format works)
3636 | uv run client.py C:/projects/mcp-server/weather.py
3637 | uv run client.py C:\\projects\\mcp-server\\weather.py
3638 | ```
3639 |
3640 | ### Response Timing
3641 |
3642 | * The first response might take up to 30 seconds to return
3643 | * This is normal and happens while:
3644 | * The server initializes
3645 | * Claude processes the query
3646 | * Tools are being executed
3647 | * Subsequent responses are typically faster
3648 | * Don't interrupt the process during this initial waiting period
3649 |
3650 | ### Common Error Messages
3651 |
3652 | If you see:
3653 |
3654 | * `FileNotFoundError`: Check your server path
3655 | * `Connection refused`: Ensure the server is running and the path is correct
3656 | * `Tool execution failed`: Verify the tool's required environment variables are set
3657 | * `Timeout error`: Consider increasing the timeout in your client configuration
3658 |
3659 | </Tab>
3660 |
3661 | <Tab title="Java">
3662 | <Note>
3663 | This is a quickstart demo based on Spring AI MCP auto-configuration and boot starters.
3664 | To learn how to create sync and async MCP Clients manually, consult the [Java SDK Client](/sdk/java/mcp-client) documentation
3665 | </Note>
3666 |
3667 | 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.
3668 | [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)
3669 |
3670 | ## System Requirements
3671 |
3672 | Before starting, ensure your system meets these requirements:
3673 |
3674 | * Java 17 or higher
3675 | * Maven 3.6+
3676 | * npx package manager
3677 | * Anthropic API key (Claude)
3678 | * Brave Search API key
3679 |
3680 | ## Setting Up Your Environment
3681 |
3682 | 1. Install npx (Node Package eXecute):
3683 | First, make sure to install [npm](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm)
3684 | and then run:
3685 | ```bash
3686 | npm install -g npx
3687 | ```
3688 |
3689 | 2. Clone the repository:
3690 | ```bash
3691 | git clone https://github.com/spring-projects/spring-ai-examples.git
3692 | cd model-context-protocol/brave-chatbot
3693 | ```
3694 |
3695 | 3. Set up your API keys:
3696 | ```bash
3697 | export ANTHROPIC_API_KEY='your-anthropic-api-key-here'
3698 | export BRAVE_API_KEY='your-brave-api-key-here'
3699 | ```
3700 |
3701 | 4. Build the application:
3702 | ```bash
3703 | ./mvnw clean install
3704 | ```
3705 |
3706 | 5. Run the application using Maven:
3707 | ```bash
3708 | ./mvnw spring-boot:run
3709 | ```
3710 |
3711 | <Warning>
3712 | Make sure you keep your `ANTHROPIC_API_KEY` and `BRAVE_API_KEY` keys secure!
3713 | </Warning>
3714 |
3715 | ## How it Works
3716 |
3717 | The application integrates Spring AI with the Brave Search MCP server through several components:
3718 |
3719 | ### MCP Client Configuration
3720 |
3721 | 1. Required dependencies in pom.xml:
3722 |
3723 | ```xml
3724 | <dependency>
3725 | <groupId>org.springframework.ai</groupId>
3726 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
3727 | </dependency>
3728 | <dependency>
3729 | <groupId>org.springframework.ai</groupId>
3730 | <artifactId>spring-ai-anthropic-spring-boot-starter</artifactId>
3731 | </dependency>
3732 | ```
3733 |
3734 | 2. Application properties (application.yml):
3735 |
3736 | ```yml
3737 | spring:
3738 | ai:
3739 | mcp:
3740 | client:
3741 | enabled: true
3742 | name: brave-search-client
3743 | version: 1.0.0
3744 | type: SYNC
3745 | request-timeout: 20s
3746 | stdio:
3747 | root-change-notification: true
3748 | servers-configuration: classpath:/mcp-servers-config.json
3749 | anthropic:
3750 | api-key: ${ANTHROPIC_API_KEY}
3751 | ```
3752 |
3753 | This activates the `spring-ai-mcp-client-spring-boot-starter` to create one or more `McpClient`s based on the provided server configuration.
3754 |
3755 | 3. MCP Server Configuration (`mcp-servers-config.json`):
3756 |
3757 | ```json
3758 | {
3759 | "mcpServers": {
3760 | "brave-search": {
3761 | "command": "npx",
3762 | "args": [
3763 | "-y",
3764 | "@modelcontextprotocol/server-brave-search"
3765 | ],
3766 | "env": {
3767 | "BRAVE_API_KEY": "<PUT YOUR BRAVE API KEY>"
3768 | }
3769 | }
3770 | }
3771 | }
3772 | ```
3773 |
3774 | ### Chat Implementation
3775 |
3776 | The chatbot is implemented using Spring AI's ChatClient with MCP tool integration:
3777 |
3778 | ```java
3779 | var chatClient = chatClientBuilder
3780 | .defaultSystem("You are useful assistant, expert in AI and Java.")
3781 | .defaultTools((Object[]) mcpToolAdapter.toolCallbacks())
3782 | .defaultAdvisors(new MessageChatMemoryAdvisor(new InMemoryChatMemory()))
3783 | .build();
3784 | ```
3785 |
3786 | Key features:
3787 |
3788 | * Uses Claude AI model for natural language understanding
3789 | * Integrates Brave Search through MCP for real-time web search capabilities
3790 | * Maintains conversation memory using InMemoryChatMemory
3791 | * Runs as an interactive command-line application
3792 |
3793 | ### Build and run
3794 |
3795 | ```bash
3796 | ./mvnw clean install
3797 | java -jar ./target/ai-mcp-brave-chatbot-0.0.1-SNAPSHOT.jar
3798 | ```
3799 |
3800 | or
3801 |
3802 | ```bash
3803 | ./mvnw spring-boot:run
3804 | ```
3805 |
3806 | 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.
3807 |
3808 | The chatbot can:
3809 |
3810 | * Answer questions using its built-in knowledge
3811 | * Perform web searches when needed using Brave Search
3812 | * Remember context from previous messages in the conversation
3813 | * Combine information from multiple sources to provide comprehensive answers
3814 |
3815 | ### Advanced Configuration
3816 |
3817 | The MCP client supports additional configuration options:
3818 |
3819 | * Client customization through `McpSyncClientCustomizer` or `McpAsyncClientCustomizer`
3820 | * Multiple clients with multiple transport types: `STDIO` and `SSE` (Server-Sent Events)
3821 | * Integration with Spring AI's tool execution framework
3822 | * Automatic client initialization and lifecycle management
3823 |
3824 | For WebFlux-based applications, you can use the WebFlux starter instead:
3825 |
3826 | ```xml
3827 | <dependency>
3828 | <groupId>org.springframework.ai</groupId>
3829 | <artifactId>spring-ai-mcp-client-webflux-spring-boot-starter</artifactId>
3830 | </dependency>
3831 | ```
3832 |
3833 | This provides similar functionality but uses a WebFlux-based SSE transport implementation, recommended for production deployments.
3834 |
3835 | </Tab>
3836 | </Tabs>
3837 |
3838 | ## Next steps
3839 |
3840 | <CardGroup cols={2}>
3841 | <Card title="Example servers" icon="grid" href="/examples">
3842 | Check out our gallery of official MCP servers and implementations
3843 | </Card>
3844 |
3845 | <Card title="Clients" icon="cubes" href="/clients">
3846 | View the list of clients that support MCP integrations
3847 | </Card>
3848 |
3849 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
3850 | Learn how to use LLMs like Claude to speed up your MCP development
3851 | </Card>
3852 |
3853 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3854 | Understand how MCP connects clients, servers, and LLMs
3855 | </Card>
3856 | </CardGroup>
3857 |
3858 | # For Server Developers
3859 |
3860 | Source: https://modelcontextprotocol.io/quickstart/server
3861 |
3862 | Get started building your own server to use in Claude for Desktop and other clients.
3863 |
3864 | 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.
3865 |
3866 | ### What we'll be building
3867 |
3868 | 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!
3869 |
3870 | 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):
3871 |
3872 | <Frame>
3873 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
3874 | </Frame>
3875 |
3876 | <Frame>
3877 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
3878 | </Frame>
3879 |
3880 | <Note>
3881 | 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).
3882 | </Note>
3883 |
3884 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
3885 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
3886 | </Accordion>
3887 |
3888 | ### Core MCP Concepts
3889 |
3890 | MCP servers can provide three main types of capabilities:
3891 |
3892 | 1. **Resources**: File-like data that can be read by clients (like API responses or file contents)
3893 | 2. **Tools**: Functions that can be called by the LLM (with user approval)
3894 | 3. **Prompts**: Pre-written templates that help users accomplish specific tasks
3895 |
3896 | This tutorial will primarily focus on tools.
3897 |
3898 | <Tabs>
3899 | <Tab title="Python">
3900 | 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)
3901 |
3902 | ### Prerequisite knowledge
3903 |
3904 | This quickstart assumes you have familiarity with:
3905 |
3906 | * Python
3907 | * LLMs like Claude
3908 |
3909 | ### System requirements
3910 |
3911 | * Python 3.10 or higher installed.
3912 | * You must use the Python MCP SDK 1.2.0 or higher.
3913 |
3914 | ### Set up your environment
3915 |
3916 | First, let's install `uv` and set up our Python project and environment:
3917 |
3918 | <CodeGroup>
3919 | ```bash MacOS/Linux
3920 | curl -LsSf https://astral.sh/uv/install.sh | sh
3921 | ```
3922 |
3923 | ```powershell Windows
3924 | powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
3925 | ```
3926 | </CodeGroup>
3927 |
3928 | Make sure to restart your terminal afterwards to ensure that the `uv` command gets picked up.
3929 |
3930 | Now, let's create and set up our project:
3931 |
3932 | <CodeGroup>
3933 | ```bash MacOS/Linux
3934 | # Create a new directory for our project
3935 | uv init weather
3936 | cd weather
3937 |
3938 | # Create virtual environment and activate it
3939 | uv venv
3940 | source .venv/bin/activate
3941 |
3942 | # Install dependencies
3943 | uv add "mcp[cli]" httpx
3944 |
3945 | # Create our server file
3946 | touch weather.py
3947 | ```
3948 |
3949 | ```powershell Windows
3950 | # Create a new directory for our project
3951 | uv init weather
3952 | cd weather
3953 |
3954 | # Create virtual environment and activate it
3955 | uv venv
3956 | .venv\Scripts\activate
3957 |
3958 | # Install dependencies
3959 | uv add mcp[cli] httpx
3960 |
3961 | # Create our server file
3962 | new-item weather.py
3963 | ```
3964 | </CodeGroup>
3965 |
3966 | Now let's dive into building your server.
3967 |
3968 | ## Building your server
3969 |
3970 | ### Importing packages and setting up the instance
3971 |
3972 | Add these to the top of your `weather.py`:
3973 |
3974 | ```python
3975 | from typing import Any
3976 | import httpx
3977 | from mcp.server.fastmcp import FastMCP
3978 |
3979 | # Initialize FastMCP server
3980 | mcp = FastMCP("weather")
3981 |
3982 | # Constants
3983 | NWS_API_BASE = "https://api.weather.gov"
3984 | USER_AGENT = "weather-app/1.0"
3985 | ```
3986 |
3987 | The FastMCP class uses Python type hints and docstrings to automatically generate tool definitions, making it easy to create and maintain MCP tools.
3988 |
3989 | ### Helper functions
3990 |
3991 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
3992 |
3993 | ```python
3994 | async def make_nws_request(url: str) -> dict[str, Any] | None:
3995 | """Make a request to the NWS API with proper error handling."""
3996 | headers = {
3997 | "User-Agent": USER_AGENT,
3998 | "Accept": "application/geo+json"
3999 | }
4000 | async with httpx.AsyncClient() as client:
4001 | try:
4002 | response = await client.get(url, headers=headers, timeout=30.0)
4003 | response.raise_for_status()
4004 | return response.json()
4005 | except Exception:
4006 | return None
4007 |
4008 | def format_alert(feature: dict) -> str:
4009 | """Format an alert feature into a readable string."""
4010 | props = feature["properties"]
4011 | return f"""
4012 | Event: {props.get('event', 'Unknown')}
4013 | Area: {props.get('areaDesc', 'Unknown')}
4014 | Severity: {props.get('severity', 'Unknown')}
4015 | Description: {props.get('description', 'No description available')}
4016 | Instructions: {props.get('instruction', 'No specific instructions provided')}
4017 | """
4018 | ```
4019 |
4020 | ### Implementing tool execution
4021 |
4022 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4023 |
4024 | ```python
4025 | @mcp.tool()
4026 | async def get_alerts(state: str) -> str:
4027 | """Get weather alerts for a US state.
4028 |
4029 | Args:
4030 | state: Two-letter US state code (e.g. CA, NY)
4031 | """
4032 | url = f"{NWS_API_BASE}/alerts/active/area/{state}"
4033 | data = await make_nws_request(url)
4034 |
4035 | if not data or "features" not in data:
4036 | return "Unable to fetch alerts or no alerts found."
4037 |
4038 | if not data["features"]:
4039 | return "No active alerts for this state."
4040 |
4041 | alerts = [format_alert(feature) for feature in data["features"]]
4042 | return "\n---\n".join(alerts)
4043 |
4044 | @mcp.tool()
4045 | async def get_forecast(latitude: float, longitude: float) -> str:
4046 | """Get weather forecast for a location.
4047 |
4048 | Args:
4049 | latitude: Latitude of the location
4050 | longitude: Longitude of the location
4051 | """
4052 | # First get the forecast grid endpoint
4053 | points_url = f"{NWS_API_BASE}/points/{latitude},{longitude}"
4054 | points_data = await make_nws_request(points_url)
4055 |
4056 | if not points_data:
4057 | return "Unable to fetch forecast data for this location."
4058 |
4059 | # Get the forecast URL from the points response
4060 | forecast_url = points_data["properties"]["forecast"]
4061 | forecast_data = await make_nws_request(forecast_url)
4062 |
4063 | if not forecast_data:
4064 | return "Unable to fetch detailed forecast."
4065 |
4066 | # Format the periods into a readable forecast
4067 | periods = forecast_data["properties"]["periods"]
4068 | forecasts = []
4069 | for period in periods[:5]: # Only show next 5 periods
4070 | forecast = f"""
4071 | {period['name']}:
4072 | Temperature: {period['temperature']}°{period['temperatureUnit']}
4073 | Wind: {period['windSpeed']} {period['windDirection']}
4074 | Forecast: {period['detailedForecast']}
4075 | """
4076 | forecasts.append(forecast)
4077 |
4078 | return "\n---\n".join(forecasts)
4079 | ```
4080 |
4081 | ### Running the server
4082 |
4083 | Finally, let's initialize and run the server:
4084 |
4085 | ```python
4086 | if __name__ == "__main__":
4087 | # Initialize and run the server
4088 | mcp.run(transport='stdio')
4089 | ```
4090 |
4091 | Your server is complete! Run `uv run weather.py` to confirm that everything's working.
4092 |
4093 | Let's now test your server from an existing MCP host, Claude for Desktop.
4094 |
4095 | ## Testing your server with Claude for Desktop
4096 |
4097 | <Note>
4098 | 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.
4099 | </Note>
4100 |
4101 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4102 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4103 |
4104 | 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.
4105 |
4106 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4107 |
4108 | <Tabs>
4109 | <Tab title="MacOS/Linux">
4110 | ```bash
4111 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4112 | ```
4113 | </Tab>
4114 |
4115 | <Tab title="Windows">
4116 | ```powershell
4117 | code $env:AppData\Claude\claude_desktop_config.json
4118 | ```
4119 | </Tab>
4120 | </Tabs>
4121 |
4122 | 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.
4123 |
4124 | In this case, we'll add our single weather server like so:
4125 |
4126 | <Tabs>
4127 | <Tab title="MacOS/Linux">
4128 | ```json Python
4129 | {
4130 | "mcpServers": {
4131 | "weather": {
4132 | "command": "uv",
4133 | "args": [
4134 | "--directory",
4135 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather",
4136 | "run",
4137 | "weather.py"
4138 | ]
4139 | }
4140 | }
4141 | }
4142 | ```
4143 | </Tab>
4144 |
4145 | <Tab title="Windows">
4146 | ```json Python
4147 | {
4148 | "mcpServers": {
4149 | "weather": {
4150 | "command": "uv",
4151 | "args": [
4152 | "--directory",
4153 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather",
4154 | "run",
4155 | "weather.py"
4156 | ]
4157 | }
4158 | }
4159 | }
4160 | ```
4161 | </Tab>
4162 | </Tabs>
4163 |
4164 | <Warning>
4165 | 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.
4166 | </Warning>
4167 |
4168 | <Note>
4169 | Make sure you pass in the absolute path to your server.
4170 | </Note>
4171 |
4172 | This tells Claude for Desktop:
4173 |
4174 | 1. There's an MCP server named "weather"
4175 | 2. To launch it by running `uv --directory /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather run weather`
4176 |
4177 | Save the file, and restart **Claude for Desktop**.
4178 |
4179 | </Tab>
4180 |
4181 | <Tab title="Node">
4182 | 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)
4183 |
4184 | ### Prerequisite knowledge
4185 |
4186 | This quickstart assumes you have familiarity with:
4187 |
4188 | * TypeScript
4189 | * LLMs like Claude
4190 |
4191 | ### System requirements
4192 |
4193 | For TypeScript, make sure you have the latest version of Node installed.
4194 |
4195 | ### Set up your environment
4196 |
4197 | First, let's install Node.js and npm if you haven't already. You can download them from [nodejs.org](https://nodejs.org/).
4198 | Verify your Node.js installation:
4199 |
4200 | ```bash
4201 | node --version
4202 | npm --version
4203 | ```
4204 |
4205 | For this tutorial, you'll need Node.js version 16 or higher.
4206 |
4207 | Now, let's create and set up our project:
4208 |
4209 | <CodeGroup>
4210 | ```bash MacOS/Linux
4211 | # Create a new directory for our project
4212 | mkdir weather
4213 | cd weather
4214 |
4215 | # Initialize a new npm project
4216 | npm init -y
4217 |
4218 | # Install dependencies
4219 | npm install @modelcontextprotocol/sdk zod
4220 | npm install -D @types/node typescript
4221 |
4222 | # Create our files
4223 | mkdir src
4224 | touch src/index.ts
4225 | ```
4226 |
4227 | ```powershell Windows
4228 | # Create a new directory for our project
4229 | md weather
4230 | cd weather
4231 |
4232 | # Initialize a new npm project
4233 | npm init -y
4234 |
4235 | # Install dependencies
4236 | npm install @modelcontextprotocol/sdk zod
4237 | npm install -D @types/node typescript
4238 |
4239 | # Create our files
4240 | md src
4241 | new-item src\index.ts
4242 | ```
4243 | </CodeGroup>
4244 |
4245 | Update your package.json to add type: "module" and a build script:
4246 |
4247 | ```json package.json
4248 | {
4249 | "type": "module",
4250 | "bin": {
4251 | "weather": "./build/index.js"
4252 | },
4253 | "scripts": {
4254 | "build": "tsc && node -e \"require('fs').chmodSync('build/index.js', '755')\"",
4255 | },
4256 | "files": [
4257 | "build"
4258 | ],
4259 | }
4260 | ```
4261 |
4262 | Create a `tsconfig.json` in the root of your project:
4263 |
4264 | ```json tsconfig.json
4265 | {
4266 | "compilerOptions": {
4267 | "target": "ES2022",
4268 | "module": "Node16",
4269 | "moduleResolution": "Node16",
4270 | "outDir": "./build",
4271 | "rootDir": "./src",
4272 | "strict": true,
4273 | "esModuleInterop": true,
4274 | "skipLibCheck": true,
4275 | "forceConsistentCasingInFileNames": true
4276 | },
4277 | "include": ["src/**/*"],
4278 | "exclude": ["node_modules"]
4279 | }
4280 | ```
4281 |
4282 | Now let's dive into building your server.
4283 |
4284 | ## Building your server
4285 |
4286 | ### Importing packages and setting up the instance
4287 |
4288 | Add these to the top of your `src/index.ts`:
4289 |
4290 | ```typescript
4291 | import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
4292 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
4293 | import { z } from "zod";
4294 |
4295 | const NWS_API_BASE = "https://api.weather.gov";
4296 | const USER_AGENT = "weather-app/1.0";
4297 |
4298 | // Create server instance
4299 | const server = new McpServer({
4300 | name: "weather",
4301 | version: "1.0.0",
4302 | });
4303 | ```
4304 |
4305 | ### Helper functions
4306 |
4307 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
4308 |
4309 | ```typescript
4310 | // Helper function for making NWS API requests
4311 | async function makeNWSRequest<T>(url: string): Promise<T | null> {
4312 | const headers = {
4313 | "User-Agent": USER_AGENT,
4314 | Accept: "application/geo+json",
4315 | };
4316 |
4317 | try {
4318 | const response = await fetch(url, { headers });
4319 | if (!response.ok) {
4320 | throw new Error(`HTTP error! status: ${response.status}`);
4321 | }
4322 | return (await response.json()) as T;
4323 | } catch (error) {
4324 | console.error("Error making NWS request:", error);
4325 | return null;
4326 | }
4327 | }
4328 |
4329 | interface AlertFeature {
4330 | properties: {
4331 | event?: string;
4332 | areaDesc?: string;
4333 | severity?: string;
4334 | status?: string;
4335 | headline?: string;
4336 | };
4337 | }
4338 |
4339 | // Format alert data
4340 | function formatAlert(feature: AlertFeature): string {
4341 | const props = feature.properties;
4342 | return [
4343 | `Event: ${props.event || "Unknown"}`,
4344 | `Area: ${props.areaDesc || "Unknown"}`,
4345 | `Severity: ${props.severity || "Unknown"}`,
4346 | `Status: ${props.status || "Unknown"}`,
4347 | `Headline: ${props.headline || "No headline"}`,
4348 | "---",
4349 | ].join("\n");
4350 | }
4351 |
4352 | interface ForecastPeriod {
4353 | name?: string;
4354 | temperature?: number;
4355 | temperatureUnit?: string;
4356 | windSpeed?: string;
4357 | windDirection?: string;
4358 | shortForecast?: string;
4359 | }
4360 |
4361 | interface AlertsResponse {
4362 | features: AlertFeature[];
4363 | }
4364 |
4365 | interface PointsResponse {
4366 | properties: {
4367 | forecast?: string;
4368 | };
4369 | }
4370 |
4371 | interface ForecastResponse {
4372 | properties: {
4373 | periods: ForecastPeriod[];
4374 | };
4375 | }
4376 | ```
4377 |
4378 | ### Implementing tool execution
4379 |
4380 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4381 |
4382 | ```typescript
4383 | // Register weather tools
4384 | server.tool(
4385 | "get-alerts",
4386 | "Get weather alerts for a state",
4387 | {
4388 | state: z.string().length(2).describe("Two-letter state code (e.g. CA, NY)"),
4389 | },
4390 | async ({ state }) => {
4391 | const stateCode = state.toUpperCase();
4392 | const alertsUrl = `${NWS_API_BASE}/alerts?area=${stateCode}`;
4393 | const alertsData = await makeNWSRequest<AlertsResponse>(alertsUrl);
4394 |
4395 | if (!alertsData) {
4396 | return {
4397 | content: [
4398 | {
4399 | type: "text",
4400 | text: "Failed to retrieve alerts data",
4401 | },
4402 | ],
4403 | };
4404 | }
4405 |
4406 | const features = alertsData.features || [];
4407 | if (features.length === 0) {
4408 | return {
4409 | content: [
4410 | {
4411 | type: "text",
4412 | text: `No active alerts for ${stateCode}`,
4413 | },
4414 | ],
4415 | };
4416 | }
4417 |
4418 | const formattedAlerts = features.map(formatAlert);
4419 | const alertsText = `Active alerts for ${stateCode}:\n\n${formattedAlerts.join("\n")}`;
4420 |
4421 | return {
4422 | content: [
4423 | {
4424 | type: "text",
4425 | text: alertsText,
4426 | },
4427 | ],
4428 | };
4429 | },
4430 | );
4431 |
4432 | server.tool(
4433 | "get-forecast",
4434 | "Get weather forecast for a location",
4435 | {
4436 | latitude: z.number().min(-90).max(90).describe("Latitude of the location"),
4437 | longitude: z.number().min(-180).max(180).describe("Longitude of the location"),
4438 | },
4439 | async ({ latitude, longitude }) => {
4440 | // Get grid point data
4441 | const pointsUrl = `${NWS_API_BASE}/points/${latitude.toFixed(4)},${longitude.toFixed(4)}`;
4442 | const pointsData = await makeNWSRequest<PointsResponse>(pointsUrl);
4443 |
4444 | if (!pointsData) {
4445 | return {
4446 | content: [
4447 | {
4448 | type: "text",
4449 | text: `Failed to retrieve grid point data for coordinates: ${latitude}, ${longitude}. This location may not be supported by the NWS API (only US locations are supported).`,
4450 | },
4451 | ],
4452 | };
4453 | }
4454 |
4455 | const forecastUrl = pointsData.properties?.forecast;
4456 | if (!forecastUrl) {
4457 | return {
4458 | content: [
4459 | {
4460 | type: "text",
4461 | text: "Failed to get forecast URL from grid point data",
4462 | },
4463 | ],
4464 | };
4465 | }
4466 |
4467 | // Get forecast data
4468 | const forecastData = await makeNWSRequest<ForecastResponse>(forecastUrl);
4469 | if (!forecastData) {
4470 | return {
4471 | content: [
4472 | {
4473 | type: "text",
4474 | text: "Failed to retrieve forecast data",
4475 | },
4476 | ],
4477 | };
4478 | }
4479 |
4480 | const periods = forecastData.properties?.periods || [];
4481 | if (periods.length === 0) {
4482 | return {
4483 | content: [
4484 | {
4485 | type: "text",
4486 | text: "No forecast periods available",
4487 | },
4488 | ],
4489 | };
4490 | }
4491 |
4492 | // Format forecast periods
4493 | const formattedForecast = periods.map((period: ForecastPeriod) =>
4494 | [
4495 | `${period.name || "Unknown"}:`,
4496 | `Temperature: ${period.temperature || "Unknown"}°${period.temperatureUnit || "F"}`,
4497 | `Wind: ${period.windSpeed || "Unknown"} ${period.windDirection || ""}`,
4498 | `${period.shortForecast || "No forecast available"}`,
4499 | "---",
4500 | ].join("\n"),
4501 | );
4502 |
4503 | const forecastText = `Forecast for ${latitude}, ${longitude}:\n\n${formattedForecast.join("\n")}`;
4504 |
4505 | return {
4506 | content: [
4507 | {
4508 | type: "text",
4509 | text: forecastText,
4510 | },
4511 | ],
4512 | };
4513 | },
4514 | );
4515 | ```
4516 |
4517 | ### Running the server
4518 |
4519 | Finally, implement the main function to run the server:
4520 |
4521 | ```typescript
4522 | async function main() {
4523 | const transport = new StdioServerTransport();
4524 | await server.connect(transport);
4525 | console.error("Weather MCP Server running on stdio");
4526 | }
4527 |
4528 | main().catch((error) => {
4529 | console.error("Fatal error in main():", error);
4530 | process.exit(1);
4531 | });
4532 | ```
4533 |
4534 | Make sure to run `npm run build` to build your server! This is a very important step in getting your server to connect.
4535 |
4536 | Let's now test your server from an existing MCP host, Claude for Desktop.
4537 |
4538 | ## Testing your server with Claude for Desktop
4539 |
4540 | <Note>
4541 | 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.
4542 | </Note>
4543 |
4544 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4545 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4546 |
4547 | 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.
4548 |
4549 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4550 |
4551 | <Tabs>
4552 | <Tab title="MacOS/Linux">
4553 | ```bash
4554 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4555 | ```
4556 | </Tab>
4557 |
4558 | <Tab title="Windows">
4559 | ```powershell
4560 | code $env:AppData\Claude\claude_desktop_config.json
4561 | ```
4562 | </Tab>
4563 | </Tabs>
4564 |
4565 | 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.
4566 |
4567 | In this case, we'll add our single weather server like so:
4568 |
4569 | <Tabs>
4570 | <Tab title="MacOS/Linux">
4571 | <CodeGroup>
4572 | ```json Node
4573 | {
4574 | "mcpServers": {
4575 | "weather": {
4576 | "command": "node",
4577 | "args": [
4578 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js"
4579 | ]
4580 | }
4581 | }
4582 | }
4583 | ```
4584 | </CodeGroup>
4585 | </Tab>
4586 |
4587 | <Tab title="Windows">
4588 | <CodeGroup>
4589 | ```json Node
4590 | {
4591 | "mcpServers": {
4592 | "weather": {
4593 | "command": "node",
4594 | "args": [
4595 | "C:\\PATH\\TO\\PARENT\\FOLDER\\weather\\build\\index.js"
4596 | ]
4597 | }
4598 | }
4599 | }
4600 | ```
4601 | </CodeGroup>
4602 | </Tab>
4603 | </Tabs>
4604 |
4605 | This tells Claude for Desktop:
4606 |
4607 | 1. There's an MCP server named "weather"
4608 | 2. Launch it by running `node /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js`
4609 |
4610 | Save the file, and restart **Claude for Desktop**.
4611 |
4612 | </Tab>
4613 |
4614 | <Tab title="Java">
4615 | <Note>
4616 | This is a quickstart demo based on Spring AI MCP auto-configuraiton and boot starters.
4617 | To learn how to create sync and async MCP Servers, manually, consult the [Java SDK Server](/sdk/java/mcp-server) documentation.
4618 | </Note>
4619 |
4620 | Let's get started with building our weather server!
4621 | [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)
4622 |
4623 | 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.
4624 | For manual MCP Server implementation, refer to the [MCP Server Java SDK documentation](/sdk/java/mcp-server).
4625 |
4626 | ### System requirements
4627 |
4628 | * Java 17 or higher installed.
4629 | * [Spring Boot 3.3.x](https://docs.spring.io/spring-boot/installing.html) or higher
4630 |
4631 | ### Set up your environment
4632 |
4633 | Use the [Spring Initizer](https://start.spring.io/) to bootstrat the project.
4634 |
4635 | You will need to add the following dependencies:
4636 |
4637 | <Tabs>
4638 | <Tab title="Maven">
4639 | ```xml
4640 | <dependencies>
4641 | <dependency>
4642 | <groupId>org.springframework.ai</groupId>
4643 | <artifactId>spring-ai-mcp-server-spring-boot-starter</artifactId>
4644 | </dependency>
4645 |
4646 | <dependency>
4647 | <groupId>org.springframework</groupId>
4648 | <artifactId>spring-web</artifactId>
4649 | </dependency>
4650 | </dependencies>
4651 | ```
4652 | </Tab>
4653 |
4654 | <Tab title="Gradle">
4655 | ```groovy
4656 | dependencies {
4657 | implementation platform("org.springframework.ai:spring-ai-mcp-server-spring-boot-starter")
4658 | implementation platform("org.springframework:spring-web")
4659 | }
4660 | ```
4661 | </Tab>
4662 | </Tabs>
4663 |
4664 | Then configure your application by setting the applicaiton properties:
4665 |
4666 | <CodeGroup>
4667 | ```bash application.properties
4668 | spring.main.bannerMode=off
4669 | logging.pattern.console=
4670 | ```
4671 |
4672 | ```yaml application.yml
4673 | logging:
4674 | pattern:
4675 | console:
4676 | spring:
4677 | main:
4678 | banner-mode: off
4679 | ```
4680 | </CodeGroup>
4681 |
4682 | 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.
4683 |
4684 | Now let's dive into building your server.
4685 |
4686 | ## Building your server
4687 |
4688 | ### Weather Service
4689 |
4690 | 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:
4691 |
4692 | ```java
4693 | @Service
4694 | public class WeatherService {
4695 |
4696 | private final RestClient restClient;
4697 |
4698 | public WeatherService() {
4699 | this.restClient = RestClient.builder()
4700 | .baseUrl("https://api.weather.gov")
4701 | .defaultHeader("Accept", "application/geo+json")
4702 | .defaultHeader("User-Agent", "WeatherApiClient/1.0 ([email protected])")
4703 | .build();
4704 | }
4705 |
4706 | @Tool(description = "Get weather forecast for a specific latitude/longitude")
4707 | public String getWeatherForecastByLocation(
4708 | double latitude, // Latitude coordinate
4709 | double longitude // Longitude coordinate
4710 | ) {
4711 | // Returns detailed forecast including:
4712 | // - Temperature and unit
4713 | // - Wind speed and direction
4714 | // - Detailed forecast description
4715 | }
4716 |
4717 | @Tool(description = "Get weather alerts for a US state")
4718 | public String getAlerts(
4719 | @ToolParam(description = "Two-letter US state code (e.g. CA, NY") String state)
4720 | ) {
4721 | // Returns active alerts including:
4722 | // - Event type
4723 | // - Affected area
4724 | // - Severity
4725 | // - Description
4726 | // - Safety instructions
4727 | }
4728 |
4729 | // ......
4730 | }
4731 | ```
4732 |
4733 | The `@Service` annotation with auto-register the service in your applicaiton context.
4734 | The Spring AI `@Tool` annotation, making it easy to create and maintain MCP tools.
4735 |
4736 | The auto-configuration will automatically register these tools with the MCP server.
4737 |
4738 | ### Create your Boot Applicaiton
4739 |
4740 | ```java
4741 | @SpringBootApplication
4742 | public class McpServerApplication {
4743 |
4744 | public static void main(String[] args) {
4745 | SpringApplication.run(McpServerApplication.class, args);
4746 | }
4747 |
4748 | @Bean
4749 | public ToolCallbackProvider weatherTools(WeatherService weatherService) {
4750 | return MethodToolCallbackProvider.builder().toolObjects(weatherService).build();
4751 | }
4752 | }
4753 | ```
4754 |
4755 | Uses the the `MethodToolCallbackProvider` utils to convert the `@Tools` into actionalble callbackes used by the MCP server.
4756 |
4757 | ### Running the server
4758 |
4759 | Finally, let's build the server:
4760 |
4761 | ```bash
4762 | ./mvnw clean install
4763 | ```
4764 |
4765 | This will generate a `mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar` file within the `target` folder.
4766 |
4767 | Let's now test your server from an existing MCP host, Claude for Desktop.
4768 |
4769 | ## Testing your server with Claude for Desktop
4770 |
4771 | <Note>
4772 | Claude for Desktop is not yet available on Linux.
4773 | </Note>
4774 |
4775 | First, make sure you have Claude for Desktop installed.
4776 | [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.**
4777 |
4778 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use.
4779 | To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor.
4780 | Make sure to create the file if it doesn't exist.
4781 |
4782 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4783 |
4784 | <Tabs>
4785 | <Tab title="MacOS/Linux">
4786 | ```bash
4787 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4788 | ```
4789 | </Tab>
4790 |
4791 | <Tab title="Windows">
4792 | ```powershell
4793 | code $env:AppData\Claude\claude_desktop_config.json
4794 | ```
4795 | </Tab>
4796 | </Tabs>
4797 |
4798 | You'll then add your servers in the `mcpServers` key.
4799 | The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4800 |
4801 | In this case, we'll add our single weather server like so:
4802 |
4803 | <Tabs>
4804 | <Tab title="MacOS/Linux">
4805 | ```json java
4806 | {
4807 | "mcpServers": {
4808 | "spring-ai-mcp-weather": {
4809 | "command": "java",
4810 | "args": [
4811 | "-Dspring.ai.mcp.server.stdio=true",
4812 | "-jar",
4813 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4814 | ]
4815 | }
4816 | }
4817 | }
4818 | ```
4819 | </Tab>
4820 |
4821 | <Tab title="Windows">
4822 | ```json java
4823 | {
4824 | "mcpServers": {
4825 | "spring-ai-mcp-weather": {
4826 | "command": "java",
4827 | "args": [
4828 | "-Dspring.ai.mcp.server.transport=STDIO",
4829 | "-jar",
4830 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather\\mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4831 | ]
4832 | }
4833 | }
4834 | }
4835 | ```
4836 | </Tab>
4837 | </Tabs>
4838 |
4839 | <Note>
4840 | Make sure you pass in the absolute path to your server.
4841 | </Note>
4842 |
4843 | This tells Claude for Desktop:
4844 |
4845 | 1. There's an MCP server named "my-weather-server"
4846 | 2. To launch it by running `java -jar /ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar`
4847 |
4848 | Save the file, and restart **Claude for Desktop**.
4849 |
4850 | ## Testing your server with Java client
4851 |
4852 | ### Create a MCP Client manually
4853 |
4854 | Use the `McpClient` to connect to the server:
4855 |
4856 | ```java
4857 | var stdioParams = ServerParameters.builder("java")
4858 | .args("-jar", "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar")
4859 | .build();
4860 |
4861 | var stdioTransport = new StdioClientTransport(stdioParams);
4862 |
4863 | var mcpClient = McpClient.sync(stdioTransport).build();
4864 |
4865 | mcpClient.initialize();
4866 |
4867 | ListToolsResult toolsList = mcpClient.listTools();
4868 |
4869 | CallToolResult weather = mcpClient.callTool(
4870 | new CallToolRequest("getWeatherForecastByLocation",
4871 | Map.of("latitude", "47.6062", "longitude", "-122.3321")));
4872 |
4873 | CallToolResult alert = mcpClient.callTool(
4874 | new CallToolRequest("getAlerts", Map.of("state", "NY")));
4875 |
4876 | mcpClient.closeGracefully();
4877 | ```
4878 |
4879 | ### Use MCP Client Boot Starter
4880 |
4881 | Create a new boot starter applicaiton using the `spring-ai-mcp-client-spring-boot-starter` dependency:
4882 |
4883 | ```xml
4884 | <dependency>
4885 | <groupId>org.springframework.ai</groupId>
4886 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
4887 | </dependency>
4888 | ```
4889 |
4890 | and set the `spring.ai.mcp.client.stdio.servers-configuration` property to point to your `claude_desktop_config.json`.
4891 | You can re-use the existing Anthropic Destop configuration:
4892 |
4893 | ```properties
4894 | spring.ai.mcp.client.stdio.servers-configuration=file:PATH/TO/claude_desktop_config.json
4895 | ```
4896 |
4897 | When you stasrt your client applicaiton, the auto-configuration will create, automatically MCP clients from the claude\_desktop\_config.json.
4898 |
4899 | 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.
4900 |
4901 | ## More Java MCP Server examples
4902 |
4903 | 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.
4904 | It showcases how to define and register MCP Tools, Resources, and Prompts, using the Spring Boot's auto-configuration capabilities.
4905 |
4906 | </Tab>
4907 | </Tabs>
4908 |
4909 | ### Test with commands
4910 |
4911 | 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:
4912 |
4913 | <Frame>
4914 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/visual-indicator-mcp-tools.png" />
4915 | </Frame>
4916 |
4917 | After clicking on the hammer icon, you should see two tools listed:
4918 |
4919 | <Frame>
4920 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/available-mcp-tools.png" />
4921 | </Frame>
4922 |
4923 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
4924 |
4925 | If the hammer icon has shown up, you can now test your server by running the following commands in Claude for Desktop:
4926 |
4927 | - What's the weather in Sacramento?
4928 | - What are the active weather alerts in Texas?
4929 |
4930 | <Frame>
4931 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
4932 | </Frame>
4933 |
4934 | <Frame>
4935 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
4936 | </Frame>
4937 |
4938 | <Note>
4939 | Since this is the US National Weather service, the queries will only work for US locations.
4940 | </Note>
4941 |
4942 | ## What's happening under the hood
4943 |
4944 | When you ask a question:
4945 |
4946 | 1. The client sends your question to Claude
4947 | 2. Claude analyzes the available tools and decides which one(s) to use
4948 | 3. The client executes the chosen tool(s) through the MCP server
4949 | 4. The results are sent back to Claude
4950 | 5. Claude formulates a natural language response
4951 | 6. The response is displayed to you!
4952 |
4953 | ## Troubleshooting
4954 |
4955 | <AccordionGroup>
4956 | <Accordion title="Claude for Desktop Integration Issues">
4957 | **Getting logs from Claude for Desktop**
4958 |
4959 | Claude.app logging related to MCP is written to log files in `~/Library/Logs/Claude`:
4960 |
4961 | * `mcp.log` will contain general logging about MCP connections and connection failures.
4962 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
4963 |
4964 | You can run the following command to list recent logs and follow along with any new ones:
4965 |
4966 | ```bash
4967 | # Check Claude's logs for errors
4968 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
4969 | ```
4970 |
4971 | **Server not showing up in Claude**
4972 |
4973 | 1. Check your `claude_desktop_config.json` file syntax
4974 | 2. Make sure the path to your project is absolute and not relative
4975 | 3. Restart Claude for Desktop completely
4976 |
4977 | **Tool calls failing silently**
4978 |
4979 | If Claude attempts to use the tools but they fail:
4980 |
4981 | 1. Check Claude's logs for errors
4982 | 2. Verify your server builds and runs without errors
4983 | 3. Try restarting Claude for Desktop
4984 |
4985 | **None of this is working. What do I do?**
4986 |
4987 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
4988 |
4989 | </Accordion>
4990 |
4991 | <Accordion title="Weather API Issues">
4992 | **Error: Failed to retrieve grid point data**
4993 |
4994 | This usually means either:
4995 |
4996 | 1. The coordinates are outside the US
4997 | 2. The NWS API is having issues
4998 | 3. You're being rate limited
4999 |
5000 | Fix:
5001 |
5002 | * Verify you're using US coordinates
5003 | * Add a small delay between requests
5004 | * Check the NWS API status page
5005 |
5006 | **Error: No active alerts for \[STATE]**
5007 |
5008 | 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.
5009 |
5010 | </Accordion>
5011 | </AccordionGroup>
5012 |
5013 | <Note>
5014 | For more advanced troubleshooting, check out our guide on [Debugging MCP](/docs/tools/debugging)
5015 | </Note>
5016 |
5017 | ## Next steps
5018 |
5019 | <CardGroup cols={2}>
5020 | <Card title="Building a client" icon="outlet" href="/quickstart/client">
5021 | Learn how to build your own MCP client that can connect to your server
5022 | </Card>
5023 |
5024 | <Card title="Example servers" icon="grid" href="/examples">
5025 | Check out our gallery of official MCP servers and implementations
5026 | </Card>
5027 |
5028 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
5029 | Learn how to effectively debug MCP servers and integrations
5030 | </Card>
5031 |
5032 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
5033 | Learn how to use LLMs like Claude to speed up your MCP development
5034 | </Card>
5035 | </CardGroup>
5036 |
5037 | # For Claude Desktop Users
5038 |
5039 | Source: https://modelcontextprotocol.io/quickstart/user
5040 |
5041 | Get started using pre-built servers in Claude for Desktop.
5042 |
5043 | 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.
5044 |
5045 | <Frame>
5046 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-filesystem.png" />
5047 | </Frame>
5048 |
5049 | Don't worry — it will ask you for your permission before executing these actions!
5050 |
5051 | ## 1. Download Claude for Desktop
5052 |
5053 | Start by downloading [Claude for Desktop](https://claude.ai/download), choosing either macOS or Windows. (Linux is not yet supported for Claude for Desktop.)
5054 |
5055 | Follow the installation instructions.
5056 |
5057 | 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..."
5058 |
5059 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
5060 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
5061 | </Accordion>
5062 |
5063 | ## 2. Add the Filesystem MCP Server
5064 |
5065 | 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.
5066 |
5067 | 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.
5068 |
5069 | This is what it should look like on a Mac:
5070 |
5071 | <Frame style={{ textAlign: 'center' }}>
5072 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-menu.png" width="400" />
5073 | </Frame>
5074 |
5075 | Click on "Developer" in the lefthand bar of the Settings pane, and then click on "Edit Config":
5076 |
5077 | <Frame>
5078 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-developer.png" />
5079 | </Frame>
5080 |
5081 | This will create a configuration file at:
5082 |
5083 | - macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
5084 | - Windows: `%APPDATA%\Claude\claude_desktop_config.json`
5085 |
5086 | if you don't already have one, and will display the file in your file system.
5087 |
5088 | Open up the configuration file in any text editor. Replace the file contents with this:
5089 |
5090 | <Tabs>
5091 | <Tab title="MacOS/Linux">
5092 | ```json
5093 | {
5094 | "mcpServers": {
5095 | "filesystem": {
5096 | "command": "npx",
5097 | "args": [
5098 | "-y",
5099 | "@modelcontextprotocol/server-filesystem",
5100 | "/Users/username/Desktop",
5101 | "/Users/username/Downloads"
5102 | ]
5103 | }
5104 | }
5105 | }
5106 | ```
5107 | </Tab>
5108 |
5109 | <Tab title="Windows">
5110 | ```json
5111 | {
5112 | "mcpServers": {
5113 | "filesystem": {
5114 | "command": "npx",
5115 | "args": [
5116 | "-y",
5117 | "@modelcontextprotocol/server-filesystem",
5118 | "C:\\Users\\username\\Desktop",
5119 | "C:\\Users\\username\\Downloads"
5120 | ]
5121 | }
5122 | }
5123 | }
5124 | ```
5125 | </Tab>
5126 | </Tabs>
5127 |
5128 | 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.
5129 |
5130 | 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.
5131 |
5132 | - On macOS, open the Terminal from your Applications folder
5133 | - On Windows, press Windows + R, type "cmd", and press Enter
5134 |
5135 | Once in the command line, verify you have Node installed by entering in the following command:
5136 |
5137 | ```bash
5138 | node --version
5139 | ```
5140 |
5141 | If you get an error saying "command not found" or "node is not recognized", download Node from [nodejs.org](https://nodejs.org/).
5142 |
5143 | <Tip>
5144 | **How does the configuration file work?**
5145 |
5146 | 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.
5147 | </Tip>
5148 |
5149 | <Warning>
5150 | **Command Privileges**
5151 |
5152 | 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.
5153 | </Warning>
5154 |
5155 | ## 3. Restart Claude
5156 |
5157 | After updating your configuration file, you need to restart Claude for Desktop.
5158 |
5159 | 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:
5160 |
5161 | <Frame>
5162 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-hammer.png" />
5163 | </Frame>
5164 |
5165 | After clicking on the hammer icon, you should see the tools that come with the Filesystem MCP Server:
5166 |
5167 | <Frame style={{ textAlign: 'center' }}>
5168 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-tools.png" width="400" />
5169 | </Frame>
5170 |
5171 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
5172 |
5173 | ## 4. Try it out!
5174 |
5175 | You can now talk to Claude and ask it about your filesystem. It should know when to call the relevant tools.
5176 |
5177 | Things you might try asking Claude:
5178 |
5179 | - Can you write a poem and save it to my desktop?
5180 | - What are some work-related files in my downloads folder?
5181 | - Can you take all the images on my desktop and move them to a new folder called "Images"?
5182 |
5183 | As needed, Claude will call the relevant tools and seek your approval before taking an action:
5184 |
5185 | <Frame style={{ textAlign: 'center' }}>
5186 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-approve.png" width="500" />
5187 | </Frame>
5188 |
5189 | ## Troubleshooting
5190 |
5191 | <AccordionGroup>
5192 | <Accordion title="Server not showing up in Claude / hammer icon missing">
5193 | 1. Restart Claude for Desktop completely
5194 | 2. Check your `claude_desktop_config.json` file syntax
5195 | 3. Make sure the file paths included in `claude_desktop_config.json` are valid and that they are absolute and not relative
5196 | 4. Look at [logs](#getting-logs-from-claude-for-desktop) to see why the server is not connecting
5197 | 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:
5198 |
5199 | <Tabs>
5200 | <Tab title="MacOS/Linux">
5201 | ```bash
5202 | npx -y @modelcontextprotocol/server-filesystem /Users/username/Desktop /Users/username/Downloads
5203 | ```
5204 | </Tab>
5205 |
5206 | <Tab title="Windows">
5207 | ```bash
5208 | npx -y @modelcontextprotocol/server-filesystem C:\Users\username\Desktop C:\Users\username\Downloads
5209 | ```
5210 | </Tab>
5211 | </Tabs>
5212 |
5213 | </Accordion>
5214 |
5215 | <Accordion title="Getting logs from Claude for Desktop">
5216 | Claude.app logging related to MCP is written to log files in:
5217 |
5218 | * macOS: `~/Library/Logs/Claude`
5219 |
5220 | * Windows: `%APPDATA%\Claude\logs`
5221 |
5222 | * `mcp.log` will contain general logging about MCP connections and connection failures.
5223 |
5224 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
5225 |
5226 | 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):
5227 |
5228 | <Tabs>
5229 | <Tab title="MacOS/Linux">
5230 | ```bash
5231 | # Check Claude's logs for errors
5232 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
5233 | ```
5234 | </Tab>
5235 |
5236 | <Tab title="Windows">
5237 | ```bash
5238 | type "%APPDATA%\Claude\logs\mcp*.log"
5239 | ```
5240 | </Tab>
5241 | </Tabs>
5242 |
5243 | </Accordion>
5244 |
5245 | <Accordion title="Tool calls failing silently">
5246 | If Claude attempts to use the tools but they fail:
5247 |
5248 | 1. Check Claude's logs for errors
5249 | 2. Verify your server builds and runs without errors
5250 | 3. Try restarting Claude for Desktop
5251 |
5252 | </Accordion>
5253 |
5254 | <Accordion title="None of this is working. What do I do?">
5255 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
5256 | </Accordion>
5257 |
5258 | <Accordion title="ENOENT error and `${APPDATA}` in paths on Windows">
5259 | If your configured server fails to load, and you see within its logs an error referring to `${APPDATA}` within a path, you may need to add the expanded value of `%APPDATA%` to your `env` key in `claude_desktop_config.json`:
5260 |
5261 | ```json
5262 | {
5263 | "brave-search": {
5264 | "command": "npx",
5265 | "args": ["-y", "@modelcontextprotocol/server-brave-search"],
5266 | "env": {
5267 | "APPDATA": "C:\\Users\\user\\AppData\\Roaming\\",
5268 | "BRAVE_API_KEY": "..."
5269 | }
5270 | }
5271 | }
5272 | ```
5273 |
5274 | With this change in place, launch Claude Desktop once again.
5275 |
5276 | <Warning>
5277 | **NPM should be installed globally**
5278 |
5279 | 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:
5280 |
5281 | ```bash
5282 | npm install -g npm
5283 | ```
5284 | </Warning>
5285 |
5286 | </Accordion>
5287 | </AccordionGroup>
5288 |
5289 | ## Next steps
5290 |
5291 | <CardGroup cols={2}>
5292 | <Card title="Explore other servers" icon="grid" href="/examples">
5293 | Check out our gallery of official MCP servers and implementations
5294 | </Card>
5295 |
5296 | <Card title="Build your own server" icon="code" href="/quickstart/server">
5297 | Now build your own custom server to use in Claude for Desktop and other clients
5298 | </Card>
5299 | </CardGroup>
5300 |
5301 | # MCP Client
5302 |
5303 | Source: https://modelcontextprotocol.io/sdk/java/mcp-client
5304 |
5305 | Learn how to use the Model Context Protocol (MCP) client to interact with MCP servers
5306 |
5307 | # Model Context Protocol Client
5308 |
5309 | 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:
5310 |
5311 | - Protocol version negotiation to ensure compatibility with servers
5312 | - Capability negotiation to determine available features
5313 | - Message transport and JSON-RPC communication
5314 | - Tool discovery and execution
5315 | - Resource access and management
5316 | - Prompt system interactions
5317 | - Optional features like roots management and sampling support
5318 |
5319 | The client provides both synchronous and asynchronous APIs for flexibility in different application contexts.
5320 |
5321 | <Tabs>
5322 | <Tab title="Sync API">
5323 | ```java
5324 | // Create a sync client with custom configuration
5325 | McpSyncClient client = McpClient.sync(transport)
5326 | .requestTimeout(Duration.ofSeconds(10))
5327 | .capabilities(ClientCapabilities.builder()
5328 | .roots(true) // Enable roots capability
5329 | .sampling() // Enable sampling capability
5330 | .build())
5331 | .sampling(request -> new CreateMessageResult(response))
5332 | .build();
5333 |
5334 | // Initialize connection
5335 | client.initialize();
5336 |
5337 | // List available tools
5338 | ListToolsResult tools = client.listTools();
5339 |
5340 | // Call a tool
5341 | CallToolResult result = client.callTool(
5342 | new CallToolRequest("calculator",
5343 | Map.of("operation", "add", "a", 2, "b", 3))
5344 | );
5345 |
5346 | // List and read resources
5347 | ListResourcesResult resources = client.listResources();
5348 | ReadResourceResult resource = client.readResource(
5349 | new ReadResourceRequest("resource://uri")
5350 | );
5351 |
5352 | // List and use prompts
5353 | ListPromptsResult prompts = client.listPrompts();
5354 | GetPromptResult prompt = client.getPrompt(
5355 | new GetPromptRequest("greeting", Map.of("name", "Spring"))
5356 | );
5357 |
5358 | // Add/remove roots
5359 | client.addRoot(new Root("file:///path", "description"));
5360 | client.removeRoot("file:///path");
5361 |
5362 | // Close client
5363 | client.closeGracefully();
5364 | ```
5365 |
5366 | </Tab>
5367 |
5368 | <Tab title="Async API">
5369 | ```java
5370 | // Create an async client with custom configuration
5371 | McpAsyncClient client = McpClient.async(transport)
5372 | .requestTimeout(Duration.ofSeconds(10))
5373 | .capabilities(ClientCapabilities.builder()
5374 | .roots(true) // Enable roots capability
5375 | .sampling() // Enable sampling capability
5376 | .build())
5377 | .sampling(request -> Mono.just(new CreateMessageResult(response)))
5378 | .toolsChangeConsumer(tools -> Mono.fromRunnable(() -> {
5379 | logger.info("Tools updated: {}", tools);
5380 | }))
5381 | .resourcesChangeConsumer(resources -> Mono.fromRunnable(() -> {
5382 | logger.info("Resources updated: {}", resources);
5383 | }))
5384 | .promptsChangeConsumer(prompts -> Mono.fromRunnable(() -> {
5385 | logger.info("Prompts updated: {}", prompts);
5386 | }))
5387 | .build();
5388 |
5389 | // Initialize connection and use features
5390 | client.initialize()
5391 | .flatMap(initResult -> client.listTools())
5392 | .flatMap(tools -> {
5393 | return client.callTool(new CallToolRequest(
5394 | "calculator",
5395 | Map.of("operation", "add", "a", 2, "b", 3)
5396 | ));
5397 | })
5398 | .flatMap(result -> {
5399 | return client.listResources()
5400 | .flatMap(resources ->
5401 | client.readResource(new ReadResourceRequest("resource://uri"))
5402 | );
5403 | })
5404 | .flatMap(resource -> {
5405 | return client.listPrompts()
5406 | .flatMap(prompts ->
5407 | client.getPrompt(new GetPromptRequest(
5408 | "greeting",
5409 | Map.of("name", "Spring")
5410 | ))
5411 | );
5412 | })
5413 | .flatMap(prompt -> {
5414 | return client.addRoot(new Root("file:///path", "description"))
5415 | .then(client.removeRoot("file:///path"));
5416 | })
5417 | .doFinally(signalType -> {
5418 | client.closeGracefully().subscribe();
5419 | })
5420 | .subscribe();
5421 | ```
5422 |
5423 | </Tab>
5424 | </Tabs>
5425 |
5426 | ## Client Transport
5427 |
5428 | 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.
5429 |
5430 | <Tabs>
5431 | <Tab title="STDIO">
5432 | Creates transport for in-process based communication
5433 |
5434 | ```java
5435 | ServerParameters params = ServerParameters.builder("npx")
5436 | .args("-y", "@modelcontextprotocol/server-everything", "dir")
5437 | .build();
5438 | McpTransport transport = new StdioClientTransport(params);
5439 | ```
5440 |
5441 | </Tab>
5442 |
5443 | <Tab title="SSE (HttpClient)">
5444 | Creates a framework agnostic (pure Java API) SSE client transport. Included in the core mcp module.
5445 |
5446 | ```java
5447 | McpTransport transport = new HttpClientSseClientTransport("http://your-mcp-server");
5448 | ```
5449 |
5450 | </Tab>
5451 |
5452 | <Tab title="SSE (WebFlux)">
5453 | Creates WebFlux-based SSE client transport. Requires the mcp-webflux-sse-transport dependency.
5454 |
5455 | ```java
5456 | WebClient.Builder webClientBuilder = WebClient.builder()
5457 | .baseUrl("http://your-mcp-server");
5458 | McpTransport transport = new WebFluxSseClientTransport(webClientBuilder);
5459 | ```
5460 |
5461 | </Tab>
5462 | </Tabs>
5463 |
5464 | ## Client Capabilities
5465 |
5466 | The client can be configured with various capabilities:
5467 |
5468 | ```java
5469 | var capabilities = ClientCapabilities.builder()
5470 | .roots(true) // Enable filesystem roots support with list changes notifications
5471 | .sampling() // Enable LLM sampling support
5472 | .build();
5473 | ```
5474 |
5475 | ### Roots Support
5476 |
5477 | Roots define the boundaries of where servers can operate within the filesystem:
5478 |
5479 | ```java
5480 | // Add a root dynamically
5481 | client.addRoot(new Root("file:///path", "description"));
5482 |
5483 | // Remove a root
5484 | client.removeRoot("file:///path");
5485 |
5486 | // Notify server of roots changes
5487 | client.rootsListChangedNotification();
5488 | ```
5489 |
5490 | The roots capability allows servers to:
5491 |
5492 | - Request the list of accessible filesystem roots
5493 | - Receive notifications when the roots list changes
5494 | - Understand which directories and files they have access to
5495 |
5496 | ### Sampling Support
5497 |
5498 | Sampling enables servers to request LLM interactions ("completions" or "generations") through the client:
5499 |
5500 | ```java
5501 | // Configure sampling handler
5502 | Function<CreateMessageRequest, CreateMessageResult> samplingHandler = request -> {
5503 | // Sampling implementation that interfaces with LLM
5504 | return new CreateMessageResult(response);
5505 | };
5506 |
5507 | // Create client with sampling support
5508 | var client = McpClient.sync(transport)
5509 | .capabilities(ClientCapabilities.builder()
5510 | .sampling()
5511 | .build())
5512 | .sampling(samplingHandler)
5513 | .build();
5514 | ```
5515 |
5516 | This capability allows:
5517 |
5518 | - Servers to leverage AI capabilities without requiring API keys
5519 | - Clients to maintain control over model access and permissions
5520 | - Support for both text and image-based interactions
5521 | - Optional inclusion of MCP server context in prompts
5522 |
5523 | ## Using MCP Clients
5524 |
5525 | ### Tool Execution
5526 |
5527 | 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.
5528 |
5529 | <Tabs>
5530 | <Tab title="Sync API">
5531 | ```java
5532 | // List available tools and their names
5533 | var tools = client.listTools();
5534 | tools.forEach(tool -> System.out.println(tool.getName()));
5535 |
5536 | // Execute a tool with parameters
5537 | var result = client.callTool("calculator", Map.of(
5538 | "operation", "add",
5539 | "a", 1,
5540 | "b", 2
5541 | ));
5542 | ```
5543 |
5544 | </Tab>
5545 |
5546 | <Tab title="Async API">
5547 | ```java
5548 | // List available tools asynchronously
5549 | client.listTools()
5550 | .doOnNext(tools -> tools.forEach(tool ->
5551 | System.out.println(tool.getName())))
5552 | .subscribe();
5553 |
5554 | // Execute a tool asynchronously
5555 | client.callTool("calculator", Map.of(
5556 | "operation", "add",
5557 | "a", 1,
5558 | "b", 2
5559 | ))
5560 | .subscribe();
5561 | ```
5562 |
5563 | </Tab>
5564 | </Tabs>
5565 |
5566 | ### Resource Access
5567 |
5568 | 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.
5569 |
5570 | <Tabs>
5571 | <Tab title="Sync API">
5572 | ```java
5573 | // List available resources and their names
5574 | var resources = client.listResources();
5575 | resources.forEach(resource -> System.out.println(resource.getName()));
5576 |
5577 | // Retrieve resource content using a URI template
5578 | var content = client.getResource("file", Map.of(
5579 | "path", "/path/to/file.txt"
5580 | ));
5581 | ```
5582 |
5583 | </Tab>
5584 |
5585 | <Tab title="Async API">
5586 | ```java
5587 | // List available resources asynchronously
5588 | client.listResources()
5589 | .doOnNext(resources -> resources.forEach(resource ->
5590 | System.out.println(resource.getName())))
5591 | .subscribe();
5592 |
5593 | // Retrieve resource content asynchronously
5594 | client.getResource("file", Map.of(
5595 | "path", "/path/to/file.txt"
5596 | ))
5597 | .subscribe();
5598 | ```
5599 |
5600 | </Tab>
5601 | </Tabs>
5602 |
5603 | ### Prompt System
5604 |
5605 | 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.
5606 |
5607 | <Tabs>
5608 | <Tab title="Sync API">
5609 | ```java
5610 | // List available prompt templates
5611 | var prompts = client.listPrompts();
5612 | prompts.forEach(prompt -> System.out.println(prompt.getName()));
5613 |
5614 | // Execute a prompt template with parameters
5615 | var response = client.executePrompt("echo", Map.of(
5616 | "text", "Hello, World!"
5617 | ));
5618 | ```
5619 |
5620 | </Tab>
5621 |
5622 | <Tab title="Async API">
5623 | ```java
5624 | // List available prompt templates asynchronously
5625 | client.listPrompts()
5626 | .doOnNext(prompts -> prompts.forEach(prompt ->
5627 | System.out.println(prompt.getName())))
5628 | .subscribe();
5629 |
5630 | // Execute a prompt template asynchronously
5631 | client.executePrompt("echo", Map.of(
5632 | "text", "Hello, World!"
5633 | ))
5634 | .subscribe();
5635 | ```
5636 |
5637 | </Tab>
5638 | </Tabs>
5639 |
5640 | # Overview
5641 |
5642 | Source: https://modelcontextprotocol.io/sdk/java/mcp-overview
5643 |
5644 | Introduction to the Model Context Protocol (MCP) Java SDK
5645 |
5646 | Java SDK for the [Model Context Protocol](https://modelcontextprotocol.org/docs/concepts/architecture)
5647 | enables standardized integration between AI models and tools.
5648 |
5649 | ## Features
5650 |
5651 | - MCP Client and MCP Server implementations supporting:
5652 | - Protocol [version compatibility negotiation](https://spec.modelcontextprotocol.io/specification/2024-11-05/basic/lifecycle/#initialization)
5653 | - [Tool](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/tools/) discovery, execution, list change notifications
5654 | - [Resource](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/resources/) management with URI templates
5655 | - [Roots](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/roots/) list management and notifications
5656 | - [Prompt](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/prompts/) handling and management
5657 | - [Sampling](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/sampling/) support for AI model interactions
5658 | - Multiple transport implementations:
5659 | - Default transports:
5660 | - Stdio-based transport for process-based communication
5661 | - Java HttpClient-based SSE client transport for HTTP SSE Client-side streaming
5662 | - Servlet-based SSE server transport for HTTP SSE Server streaming
5663 | - Spring-based transports:
5664 | - WebFlux SSE client and server transports for reactive HTTP streaming
5665 | - WebMVC SSE transport for servlet-based HTTP streaming
5666 | - Supports Synchronous and Asynchronous programming paradigms
5667 |
5668 | ## Architecture
5669 |
5670 | The SDK follows a layered architecture with clear separation of concerns:
5671 |
5672 | 
5673 |
5674 | - **Client/Server Layer (McpClient/McpServer)**: Both use McpSession for sync/async operations,
5675 | with McpClient handling client-side protocol operations and McpServer managing server-side protocol operations.
5676 | - **Session Layer (McpSession)**: Manages communication patterns and state using DefaultMcpSession implementation.
5677 | - **Transport Layer (McpTransport)**: Handles JSON-RPC message serialization/deserialization via:
5678 | - StdioTransport (stdin/stdout) in the core module
5679 | - HTTP SSE transports in dedicated transport modules (Java HttpClient, Spring WebFlux, Spring WebMVC)
5680 |
5681 | The MCP Client is a key component in the Model Context Protocol (MCP) architecture, responsible for establishing and managing connections with MCP servers.
5682 | It implements the client-side of the protocol.
5683 |
5684 | 
5685 |
5686 | The MCP Server is a foundational component in the Model Context Protocol (MCP) architecture that provides tools, resources, and capabilities to clients.
5687 | It implements the server-side of the protocol.
5688 |
5689 | 
5690 |
5691 | Key Interactions:
5692 |
5693 | - **Client/Server Initialization**: Transport setup, protocol compatibility check, capability negotiation, and implementation details exchange.
5694 | - **Message Flow**: JSON-RPC message handling with validation, type-safe response processing, and error handling.
5695 | - **Resource Management**: Resource discovery, URI template-based access, subscription system, and content retrieval.
5696 |
5697 | ## Dependencies
5698 |
5699 | Add the following Maven dependency to your project:
5700 |
5701 | <Tabs>
5702 | <Tab title="Maven">
5703 | The core MCP functionality:
5704 |
5705 | ```xml
5706 | <dependency>
5707 | <groupId>io.modelcontextprotocol.sdk</groupId>
5708 | <artifactId>mcp</artifactId>
5709 | </dependency>
5710 | ```
5711 |
5712 | For HTTP SSE transport implementations, add one of the following dependencies:
5713 |
5714 | ```xml
5715 | <!-- Spring WebFlux-based SSE client and server transport -->
5716 | <dependency>
5717 | <groupId>io.modelcontextprotocol.sdk</groupId>
5718 | <artifactId>mcp-spring-webflux</artifactId>
5719 | </dependency>
5720 |
5721 | <!-- Spring WebMVC-based SSE server transport -->
5722 | <dependency>
5723 | <groupId>io.modelcontextprotocol.sdk</groupId>
5724 | <artifactId>mcp-spring-webmvc</artifactId>
5725 | </dependency>
5726 | ```
5727 |
5728 | </Tab>
5729 |
5730 | <Tab title="Gradle">
5731 | The core MCP functionality:
5732 |
5733 | ```groovy
5734 | dependencies {
5735 | implementation platform("io.modelcontextprotocol.sdk:mcp")
5736 | //...
5737 | }
5738 | ```
5739 |
5740 | For HTTP SSE transport implementations, add one of the following dependencies:
5741 |
5742 | ```groovy
5743 | // Spring WebFlux-based SSE client and server transport
5744 | dependencies {
5745 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webflux")
5746 | }
5747 |
5748 | // Spring WebMVC-based SSE server transport
5749 | dependencies {
5750 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webmvc")
5751 | }
5752 | ```
5753 |
5754 | </Tab>
5755 | </Tabs>
5756 |
5757 | ### Bill of Materials (BOM)
5758 |
5759 | The Bill of Materials (BOM) declares the recommended versions of all the dependencies used by a given release.
5760 | Using the BOM from your application's build script avoids the need for you to specify and maintain the dependency versions yourself.
5761 | Instead, the version of the BOM you're using determines the utilized dependency versions.
5762 | It also ensures that you're using supported and tested versions of the dependencies by default, unless you choose to override them.
5763 |
5764 | Add the BOM to your project:
5765 |
5766 | <Tabs>
5767 | <Tab title="Maven">
5768 | ```xml
5769 | <dependencyManagement>
5770 | <dependencies>
5771 | <dependency>
5772 | <groupId>io.modelcontextprotocol.sdk</groupId>
5773 | <artifactId>mcp-bom</artifactId>
5774 | <version>0.7.0</version>
5775 | <type>pom</type>
5776 | <scope>import</scope>
5777 | </dependency>
5778 | </dependencies>
5779 | </dependencyManagement>
5780 | ```
5781 | </Tab>
5782 |
5783 | <Tab title="Gradle">
5784 | ```groovy
5785 | dependencies {
5786 | implementation platform("io.modelcontextprotocol.sdk:mcp-bom:0.7.0")
5787 | //...
5788 | }
5789 | ```
5790 |
5791 | 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.
5792 | This is implemented by adding a 'platform' dependency handler method to the dependencies section of your Gradle build script.
5793 | 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.
5794 |
5795 | </Tab>
5796 | </Tabs>
5797 |
5798 | Replace the version number with the version of the BOM you want to use.
5799 |
5800 | ### Available Dependencies
5801 |
5802 | The following dependencies are available and managed by the BOM:
5803 |
5804 | - Core Dependencies
5805 | - `io.modelcontextprotocol.sdk:mcp` - Core MCP library providing the base functionality and APIs for Model Context Protocol implementation.
5806 | - Transport Dependencies
5807 | - `io.modelcontextprotocol.sdk:mcp-spring-webflux` - WebFlux-based Server-Sent Events (SSE) transport implementation for reactive applications.
5808 | - `io.modelcontextprotocol.sdk:mcp-spring-webmvc` - WebMVC-based Server-Sent Events (SSE) transport implementation for servlet-based applications.
5809 | - Testing Dependencies
5810 | - `io.modelcontextprotocol.sdk:mcp-test` - Testing utilities and support for MCP-based applications.
5811 |
5812 | # MCP Server
5813 |
5814 | Source: https://modelcontextprotocol.io/sdk/java/mcp-server
5815 |
5816 | Learn how to implement and configure a Model Context Protocol (MCP) server
5817 |
5818 | ## Overview
5819 |
5820 | 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:
5821 |
5822 | - Exposing tools that clients can discover and execute
5823 | - Managing resources with URI-based access patterns
5824 | - Providing prompt templates and handling prompt requests
5825 | - Supporting capability negotiation with clients
5826 | - Implementing server-side protocol operations
5827 | - Managing concurrent client connections
5828 | - Providing structured logging and notifications
5829 |
5830 | The server supports both synchronous and asynchronous APIs, allowing for flexible integration in different application contexts.
5831 |
5832 | <Tabs>
5833 | <Tab title="Sync API">
5834 | ```java
5835 | // Create a server with custom configuration
5836 | McpSyncServer syncServer = McpServer.sync(transport)
5837 | .serverInfo("my-server", "1.0.0")
5838 | .capabilities(ServerCapabilities.builder()
5839 | .resources(true) // Enable resource support
5840 | .tools(true) // Enable tool support
5841 | .prompts(true) // Enable prompt support
5842 | .logging() // Enable logging support
5843 | .build())
5844 | .build();
5845 |
5846 | // Register tools, resources, and prompts
5847 | syncServer.addTool(syncToolRegistration);
5848 | syncServer.addResource(syncResourceRegistration);
5849 | syncServer.addPrompt(syncPromptRegistration);
5850 |
5851 | // Send logging notifications
5852 | syncServer.loggingNotification(LoggingMessageNotification.builder()
5853 | .level(LoggingLevel.INFO)
5854 | .logger("custom-logger")
5855 | .data("Server initialized")
5856 | .build());
5857 |
5858 | // Close the server when done
5859 | syncServer.close();
5860 | ```
5861 |
5862 | </Tab>
5863 |
5864 | <Tab title="Async API">
5865 | ```java
5866 | // Create an async server with custom configuration
5867 | McpAsyncServer asyncServer = McpServer.async(transport)
5868 | .serverInfo("my-server", "1.0.0")
5869 | .capabilities(ServerCapabilities.builder()
5870 | .resources(true) // Enable resource support
5871 | .tools(true) // Enable tool support
5872 | .prompts(true) // Enable prompt support
5873 | .logging() // Enable logging support
5874 | .build())
5875 | .build();
5876 |
5877 | // Register tools, resources, and prompts
5878 | asyncServer.addTool(asyncToolRegistration)
5879 | .doOnSuccess(v -> logger.info("Tool registered"))
5880 | .subscribe();
5881 |
5882 | asyncServer.addResource(asyncResourceRegistration)
5883 | .doOnSuccess(v -> logger.info("Resource registered"))
5884 | .subscribe();
5885 |
5886 | asyncServer.addPrompt(asyncPromptRegistration)
5887 | .doOnSuccess(v -> logger.info("Prompt registered"))
5888 | .subscribe();
5889 |
5890 | // Send logging notifications
5891 | asyncServer.loggingNotification(LoggingMessageNotification.builder()
5892 | .level(LoggingLevel.INFO)
5893 | .logger("custom-logger")
5894 | .data("Server initialized")
5895 | .build());
5896 |
5897 | // Close the server when done
5898 | asyncServer.close()
5899 | .doOnSuccess(v -> logger.info("Server closed"))
5900 | .subscribe();
5901 | ```
5902 |
5903 | </Tab>
5904 | </Tabs>
5905 |
5906 | ## Server Transport
5907 |
5908 | 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:
5909 |
5910 | <Tabs>
5911 | <Tab title="STDIO">
5912 | <>
5913 | Create in-process based transport:
5914 |
5915 | ```java
5916 | StdioServerTransport transport = new StdioServerTransport(new ObjectMapper());
5917 | ```
5918 |
5919 | Provides bidirectional JSON-RPC message handling over standard input/output streams with non-blocking message processing, serialization/deserialization, and graceful shutdown support.
5920 |
5921 | Key features:
5922 |
5923 | <ul>
5924 | <li>Bidirectional communication through stdin/stdout</li>
5925 | <li>Process-based integration support</li>
5926 | <li>Simple setup and configuration</li>
5927 | <li>Lightweight implementation</li>
5928 | </ul>
5929 | </>
5930 |
5931 | </Tab>
5932 |
5933 | <Tab title="SSE (WebFlux)">
5934 | <>
5935 | <p>Creates WebFlux-based SSE server transport.<br />Requires the <code>mcp-spring-webflux</code> dependency.</p>
5936 |
5937 | ```java
5938 | @Configuration
5939 | class McpConfig {
5940 | @Bean
5941 | WebFluxSseServerTransport webFluxSseServerTransport(ObjectMapper mapper) {
5942 | return new WebFluxSseServerTransport(mapper, "/mcp/message");
5943 | }
5944 |
5945 | @Bean
5946 | RouterFunction<?> mcpRouterFunction(WebFluxSseServerTransport transport) {
5947 | return transport.getRouterFunction();
5948 | }
5949 | }
5950 | ```
5951 |
5952 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5953 |
5954 | <ul>
5955 | <li>Reactive HTTP streaming with WebFlux</li>
5956 | <li>Concurrent client connections through SSE endpoints</li>
5957 | <li>Message routing and session management</li>
5958 | <li>Graceful shutdown capabilities</li>
5959 | </ul>
5960 | </>
5961 |
5962 | </Tab>
5963 |
5964 | <Tab title="SSE (WebMvc)">
5965 | <>
5966 | <p>Creates WebMvc-based SSE server transport.<br />Requires the <code>mcp-spring-webmvc</code> dependency.</p>
5967 |
5968 | ```java
5969 | @Configuration
5970 | @EnableWebMvc
5971 | class McpConfig {
5972 | @Bean
5973 | WebMvcSseServerTransport webMvcSseServerTransport(ObjectMapper mapper) {
5974 | return new WebMvcSseServerTransport(mapper, "/mcp/message");
5975 | }
5976 |
5977 | @Bean
5978 | RouterFunction<ServerResponse> mcpRouterFunction(WebMvcSseServerTransport transport) {
5979 | return transport.getRouterFunction();
5980 | }
5981 | }
5982 | ```
5983 |
5984 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5985 |
5986 | <ul>
5987 | <li>Server-side event streaming</li>
5988 | <li>Integration with Spring WebMVC</li>
5989 | <li>Support for traditional web applications</li>
5990 | <li>Synchronous operation handling</li>
5991 | </ul>
5992 | </>
5993 |
5994 | </Tab>
5995 |
5996 | <Tab title="SSE (Servlet)">
5997 | <>
5998 | <p>
5999 | Creates a Servlet-based SSE server transport. It is included in the core <code>mcp</code> module.<br />
6000 | The <code>HttpServletSseServerTransport</code> can be used with any Servlet container.<br />
6001 | To use it with a Spring Web application, you can register it as a Servlet bean:
6002 | </p>
6003 |
6004 | ```java
6005 | @Configuration
6006 | @EnableWebMvc
6007 | public class McpServerConfig implements WebMvcConfigurer {
6008 |
6009 | @Bean
6010 | public HttpServletSseServerTransport servletSseServerTransport() {
6011 | return new HttpServletSseServerTransport(new ObjectMapper(), "/mcp/message");
6012 | }
6013 |
6014 | @Bean
6015 | public ServletRegistrationBean customServletBean(HttpServletSseServerTransport servlet) {
6016 | return new ServletRegistrationBean(servlet);
6017 | }
6018 | }
6019 | ```
6020 |
6021 | <p>
6022 | Implements the MCP HTTP with SSE transport specification using the traditional Servlet API, providing:
6023 | </p>
6024 |
6025 | <ul>
6026 | <li>Asynchronous message handling using Servlet 6.0 async support</li>
6027 | <li>Session management for multiple client connections</li>
6028 |
6029 | <li>
6030 | Two types of endpoints:
6031 |
6032 | <ul>
6033 | <li>SSE endpoint (<code>/sse</code>) for server-to-client events</li>
6034 | <li>Message endpoint (configurable) for client-to-server requests</li>
6035 | </ul>
6036 | </li>
6037 |
6038 | <li>Error handling and response formatting</li>
6039 | <li>Graceful shutdown support</li>
6040 | </ul>
6041 | </>
6042 |
6043 | </Tab>
6044 | </Tabs>
6045 |
6046 | ## Server Capabilities
6047 |
6048 | The server can be configured with various capabilities:
6049 |
6050 | ```java
6051 | var capabilities = ServerCapabilities.builder()
6052 | .resources(false, true) // Resource support with list changes notifications
6053 | .tools(true) // Tool support with list changes notifications
6054 | .prompts(true) // Prompt support with list changes notifications
6055 | .logging() // Enable logging support (enabled by default with loging level INFO)
6056 | .build();
6057 | ```
6058 |
6059 | ### Logging Support
6060 |
6061 | The server provides structured logging capabilities that allow sending log messages to clients with different severity levels:
6062 |
6063 | ```java
6064 | // Send a log message to clients
6065 | server.loggingNotification(LoggingMessageNotification.builder()
6066 | .level(LoggingLevel.INFO)
6067 | .logger("custom-logger")
6068 | .data("Custom log message")
6069 | .build());
6070 | ```
6071 |
6072 | Clients can control the minimum logging level they receive through the `mcpClient.setLoggingLevel(level)` request. Messages below the set level will be filtered out.
6073 | Supported logging levels (in order of increasing severity): DEBUG (0), INFO (1), NOTICE (2), WARNING (3), ERROR (4), CRITICAL (5), ALERT (6), EMERGENCY (7)
6074 |
6075 | ### Tool Registration
6076 |
6077 | <Tabs>
6078 | <Tab title="Sync">
6079 | ```java
6080 | // Sync tool registration
6081 | var syncToolRegistration = new McpServerFeatures.SyncToolRegistration(
6082 | new Tool("calculator", "Basic calculator", Map.of(
6083 | "operation", "string",
6084 | "a", "number",
6085 | "b", "number"
6086 | )),
6087 | arguments -> {
6088 | // Tool implementation
6089 | return new CallToolResult(result, false);
6090 | }
6091 | );
6092 | ```
6093 | </Tab>
6094 |
6095 | <Tab title="Async">
6096 | ```java
6097 | // Async tool registration
6098 | var asyncToolRegistration = new McpServerFeatures.AsyncToolRegistration(
6099 | new Tool("calculator", "Basic calculator", Map.of(
6100 | "operation", "string",
6101 | "a", "number",
6102 | "b", "number"
6103 | )),
6104 | arguments -> {
6105 | // Tool implementation
6106 | return Mono.just(new CallToolResult(result, false));
6107 | }
6108 | );
6109 | ```
6110 | </Tab>
6111 | </Tabs>
6112 |
6113 | ### Resource Registration
6114 |
6115 | <Tabs>
6116 | <Tab title="Sync">
6117 | ```java
6118 | // Sync resource registration
6119 | var syncResourceRegistration = new McpServerFeatures.SyncResourceRegistration(
6120 | new Resource("custom://resource", "name", "description", "mime-type", null),
6121 | request -> {
6122 | // Resource read implementation
6123 | return new ReadResourceResult(contents);
6124 | }
6125 | );
6126 | ```
6127 | </Tab>
6128 |
6129 | <Tab title="Async">
6130 | ```java
6131 | // Async resource registration
6132 | var asyncResourceRegistration = new McpServerFeatures.AsyncResourceRegistration(
6133 | new Resource("custom://resource", "name", "description", "mime-type", null),
6134 | request -> {
6135 | // Resource read implementation
6136 | return Mono.just(new ReadResourceResult(contents));
6137 | }
6138 | );
6139 | ```
6140 | </Tab>
6141 | </Tabs>
6142 |
6143 | ### Prompt Registration
6144 |
6145 | <Tabs>
6146 | <Tab title="Sync">
6147 | ```java
6148 | // Sync prompt registration
6149 | var syncPromptRegistration = new McpServerFeatures.SyncPromptRegistration(
6150 | new Prompt("greeting", "description", List.of(
6151 | new PromptArgument("name", "description", true)
6152 | )),
6153 | request -> {
6154 | // Prompt implementation
6155 | return new GetPromptResult(description, messages);
6156 | }
6157 | );
6158 | ```
6159 | </Tab>
6160 |
6161 | <Tab title="Async">
6162 | ```java
6163 | // Async prompt registration
6164 | var asyncPromptRegistration = new McpServerFeatures.AsyncPromptRegistration(
6165 | new Prompt("greeting", "description", List.of(
6166 | new PromptArgument("name", "description", true)
6167 | )),
6168 | request -> {
6169 | // Prompt implementation
6170 | return Mono.just(new GetPromptResult(description, messages));
6171 | }
6172 | );
6173 | ```
6174 | </Tab>
6175 | </Tabs>
6176 |
6177 | ## Error Handling
6178 |
6179 | 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.
6180 |
6181 | # Building MCP with LLMs
6182 |
6183 | Source: https://modelcontextprotocol.io/tutorials/building-mcp-with-llms
6184 |
6185 | Speed up your MCP development using LLMs such as Claude!
6186 |
6187 | 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.
6188 |
6189 | ## Preparing the documentation
6190 |
6191 | Before starting, gather the necessary documentation to help Claude understand MCP:
6192 |
6193 | 1. Visit [https://modelcontextprotocol.io/llms-full.txt](https://modelcontextprotocol.io/llms-full.txt) and copy the full documentation text
6194 | 2. Navigate to either the [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) or [Python SDK repository](https://github.com/modelcontextprotocol/python-sdk)
6195 | 3. Copy the README files and other relevant documentation
6196 | 4. Paste these documents into your conversation with Claude
6197 |
6198 | ## Describing your server
6199 |
6200 | Once you've provided the documentation, clearly describe to Claude what kind of server you want to build. Be specific about:
6201 |
6202 | - What resources your server will expose
6203 | - What tools it will provide
6204 | - Any prompts it should offer
6205 | - What external systems it needs to interact with
6206 |
6207 | For example:
6208 |
6209 | ```
6210 | Build an MCP server that:
6211 | - Connects to my company's PostgreSQL database
6212 | - Exposes table schemas as resources
6213 | - Provides tools for running read-only SQL queries
6214 | - Includes prompts for common data analysis tasks
6215 | ```
6216 |
6217 | ## Working with Claude
6218 |
6219 | When working with Claude on MCP servers:
6220 |
6221 | 1. Start with the core functionality first, then iterate to add more features
6222 | 2. Ask Claude to explain any parts of the code you don't understand
6223 | 3. Request modifications or improvements as needed
6224 | 4. Have Claude help you test the server and handle edge cases
6225 |
6226 | Claude can help implement all the key MCP features:
6227 |
6228 | - Resource management and exposure
6229 | - Tool definitions and implementations
6230 | - Prompt templates and handlers
6231 | - Error handling and logging
6232 | - Connection and transport setup
6233 |
6234 | ## Best practices
6235 |
6236 | When building MCP servers with Claude:
6237 |
6238 | - Break down complex servers into smaller pieces
6239 | - Test each component thoroughly before moving on
6240 | - Keep security in mind - validate inputs and limit access appropriately
6241 | - Document your code well for future maintenance
6242 | - Follow MCP protocol specifications carefully
6243 |
6244 | ## Next steps
6245 |
6246 | After Claude helps you build your server:
6247 |
6248 | 1. Review the generated code carefully
6249 | 2. Test the server with the MCP Inspector tool
6250 | 3. Connect it to Claude.app or other MCP clients
6251 | 4. Iterate based on real usage and feedback
6252 |
6253 | Remember that Claude can help you modify and improve your server as requirements change over time.
6254 |
6255 | Need more guidance? Just ask Claude specific questions about implementing MCP features or troubleshooting issues that arise.
6256 |
```