This is page 3 of 3. Use http://codebase.md/hrgarber/wagyu_mcp_hackathon?lines=true&page={x} to view the full context.
# Directory Structure
```
├── .github
│ ├── hooks
│ │ ├── modules
│ │ │ ├── api-key-check.sh
│ │ │ └── env-check.sh
│ │ └── pre-commit
│ ├── scripts
│ │ └── auto-setup.sh
│ └── workflows
│ └── auto-setup-hooks.yml
├── .gitignore
├── assets
│ └── images
│ └── wagyu_ninja.png
├── docs
│ ├── README.md
│ └── reagan_planning
│ ├── meeting with reagan
│ │ └── transcript.md
│ └── transcript_insights.md
├── LICENSE
├── old
│ ├── docs
│ │ ├── 2025-01-10_odds_api_v4.md
│ │ ├── 2025-01-15_mcp_infra.md
│ │ ├── 2025-01-20_api_key_security.md
│ │ ├── 2025-02-10_pip_install_fix_plan.md
│ │ ├── 2025-02-15_python_odds_api_fix_postmortem.md
│ │ └── 2025-02-20_task_context.md
│ ├── espn_nonbetting_api
│ │ └── espn_api_endpoints.md
│ ├── README.md
│ └── reagan_planning
│ └── mcp_testing_approach.md
├── README.md
└── wagyu_sports
├── __init__.py
├── build
│ ├── pyproject.toml
│ ├── requirements.txt
│ └── setup.py
├── config
│ ├── .env.example
│ └── pytest.ini
├── conftest.py
├── docs
│ └── LICENSE
├── examples
│ ├── advanced_example.py
│ ├── example.py
│ ├── fetch_nba_odds.py
│ ├── verify_import.py
│ └── verify_install.py
├── Makefile
├── mcp_server
│ ├── __init__.py
│ ├── capture_live_responses.py
│ ├── mocks_live
│ │ ├── nba_games_live.json
│ │ ├── quota_info_live.json
│ │ ├── README.md
│ │ └── sports_list_live.json
│ ├── odds_client_server.py
│ ├── odds_client.py
│ ├── README.md
│ └── test_server.py
├── odds_client.py
├── README.md
├── tests
│ ├── README.md
│ ├── test_odds_api.py
│ ├── test_odds_mcp_server.py
│ └── test_simple_mcp.py
└── utils.py
```
# Files
--------------------------------------------------------------------------------
/old/docs/2025-01-15_mcp_infra.md:
--------------------------------------------------------------------------------
```markdown
1 | # Example Clients
2 | Source: https://modelcontextprotocol.io/clients
3 |
4 | A list of applications that support MCP integrations
5 |
6 | This page provides an overview of applications that support the Model Context Protocol (MCP). Each client may support different MCP features, allowing for varying levels of integration with MCP servers.
7 |
8 | ## Feature support matrix
9 |
10 | | Client | [Resources] | [Prompts] | [Tools] | [Sampling] | Roots | Notes |
11 | | ------------------------------------ | ----------- | --------- | ------- | ---------- | ----- | ------------------------------------------------------------------ |
12 | | [Claude Desktop App][Claude] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
13 | | [5ire][5ire] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
14 | | [BeeAI Framework][BeeAI Framework] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in agentic workflows. |
15 | | [Cline][Cline] | ✅ | ❌ | ✅ | ❌ | ❌ | Supports tools and resources. |
16 | | [Continue][Continue] | ✅ | ✅ | ✅ | ❌ | ❌ | Full support for all MCP features |
17 | | [Cursor][Cursor] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
18 | | [Emacs Mcp][Mcp.el] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in Emacs. |
19 | | [Firebase Genkit][Genkit] | ⚠️ | ✅ | ✅ | ❌ | ❌ | Supports resource list and lookup through tools. |
20 | | [GenAIScript][GenAIScript] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
21 | | [Goose][Goose] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools. |
22 | | [LibreChat][LibreChat] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Agents |
23 | | [mcp-agent][mcp-agent] | ❌ | ❌ | ✅ | ⚠️ | ❌ | Supports tools, server connection management, and agent workflows. |
24 | | [Roo Code][Roo Code] | ✅ | ❌ | ✅ | ❌ | ❌ | Supports tools and resources. |
25 | | [Sourcegraph Cody][Cody] | ✅ | ❌ | ❌ | ❌ | ❌ | Supports resources through OpenCTX |
26 | | [Superinterface][Superinterface] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools |
27 | | [TheiaAI/TheiaIDE][TheiaAI/TheiaIDE] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools for Agents in Theia AI and the AI-powered Theia IDE |
28 | | [Windsurf Editor][Windsurf] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools with AI Flow for collaborative development. |
29 | | [Zed][Zed] | ❌ | ✅ | ❌ | ❌ | ❌ | Prompts appear as slash commands |
30 | | \[OpenSumi]\[OpenSumi] | ❌ | ❌ | ✅ | ❌ | ❌ | Supports tools in OpenSumi |
31 |
32 | [Claude]: https://claude.ai/download
33 |
34 | [Cursor]: https://cursor.com
35 |
36 | [Zed]: https://zed.dev
37 |
38 | [Cody]: https://sourcegraph.com/cody
39 |
40 | [Genkit]: https://github.com/firebase/genkit
41 |
42 | [Continue]: https://github.com/continuedev/continue
43 |
44 | [GenAIScript]: https://microsoft.github.io/genaiscript/reference/scripts/mcp-tools/
45 |
46 | [Cline]: https://github.com/cline/cline
47 |
48 | [LibreChat]: https://github.com/danny-avila/LibreChat
49 |
50 | [TheiaAI/TheiaIDE]: https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/
51 |
52 | [Superinterface]: https://superinterface.ai
53 |
54 | [5ire]: https://github.com/nanbingxyz/5ire
55 |
56 | [BeeAI Framework]: https://i-am-bee.github.io/beeai-framework
57 |
58 | [mcp-agent]: https://github.com/lastmile-ai/mcp-agent
59 |
60 | [Mcp.el]: https://github.com/lizqwerscott/mcp.el
61 |
62 | [Roo Code]: https://roocode.com
63 |
64 | [Goose]: https://block.github.io/goose/docs/goose-architecture/#interoperability-with-extensions
65 |
66 | [Windsurf]: https://codeium.com/windsurf
67 |
68 | [Resources]: https://modelcontextprotocol.io/docs/concepts/resources
69 |
70 | [Prompts]: https://modelcontextprotocol.io/docs/concepts/prompts
71 |
72 | [Tools]: https://modelcontextprotocol.io/docs/concepts/tools
73 |
74 | [Sampling]: https://modelcontextprotocol.io/docs/concepts/sampling
75 |
76 | ## Client details
77 |
78 | ### Claude Desktop App
79 |
80 | The Claude desktop application provides comprehensive support for MCP, enabling deep integration with local tools and data sources.
81 |
82 | **Key features:**
83 |
84 | * Full support for resources, allowing attachment of local files and data
85 | * Support for prompt templates
86 | * Tool integration for executing commands and scripts
87 | * Local server connections for enhanced privacy and security
88 |
89 | > ⓘ Note: The Claude.ai web application does not currently support MCP. MCP features are only available in the desktop application.
90 |
91 | ### 5ire
92 |
93 | [5ire](https://github.com/nanbingxyz/5ire) is an open source cross-platform desktop AI assistant that supports tools through MCP servers.
94 |
95 | **Key features:**
96 |
97 | * Built-in MCP servers can be quickly enabled and disabled.
98 | * Users can add more servers by modifying the configuration file.
99 | * It is open-source and user-friendly, suitable for beginners.
100 | * Future support for MCP will be continuously improved.
101 |
102 | ### BeeAI Framework
103 |
104 | [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.
105 |
106 | **Key features:**
107 |
108 | * Seamlessly incorporate MCP tools into agentic workflows.
109 | * Quickly instantiate framework-native tools from connected MCP client(s).
110 | * Planned future support for agentic MCP capabilities.
111 |
112 | **Learn more:**
113 |
114 | * [Example of using MCP tools in agentic workflow](https://i-am-bee.github.io/beeai-framework/#/typescript/tools?id=using-the-mcptool-class)
115 |
116 | ### Cline
117 |
118 | [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.
119 |
120 | **Key features:**
121 |
122 | * Create and add tools through natural language (e.g. "add a tool that searches the web")
123 | * Share custom MCP servers Cline creates with others via the `~/Documents/Cline/MCP` directory
124 | * Displays configured MCP servers along with their tools, resources, and any error logs
125 |
126 | ### Continue
127 |
128 | [Continue](https://github.com/continuedev/continue) is an open-source AI code assistant, with built-in support for all MCP features.
129 |
130 | **Key features**
131 |
132 | * Type "@" to mention MCP resources
133 | * Prompt templates surface as slash commands
134 | * Use both built-in and MCP tools directly in chat
135 | * Supports VS Code and JetBrains IDEs, with any LLM
136 |
137 | ### Cursor
138 |
139 | [Cursor](https://docs.cursor.com/advanced/model-context-protocol) is an AI code editor.
140 |
141 | **Key Features**:
142 |
143 | * Support for MCP tools in Cursor Composer
144 | * Support for both STDIO and SSE
145 |
146 | ### Emacs Mcp
147 |
148 | [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.
149 |
150 | **Key features:**
151 |
152 | * Provides MCP tool support for Emacs.
153 |
154 | ### Firebase Genkit
155 |
156 | [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.
157 |
158 | **Key features:**
159 |
160 | * Client support for tools and prompts (resources partially supported)
161 | * Rich discovery with support in Genkit's Dev UI playground
162 | * Seamless interoperability with Genkit's existing tools and prompts
163 | * Works across a wide variety of GenAI models from top providers
164 |
165 | ### GenAIScript
166 |
167 | Programmatically assemble prompts for LLMs using [GenAIScript](https://microsoft.github.io/genaiscript/) (in JavaScript). Orchestrate LLMs, tools, and data in JavaScript.
168 |
169 | **Key features:**
170 |
171 | * JavaScript toolbox to work with prompts
172 | * Abstraction to make it easy and productive
173 | * Seamless Visual Studio Code integration
174 |
175 | ### Goose
176 |
177 | [Goose](https://github.com/block/goose) is an open source AI agent that supercharges your software development by automating coding tasks.
178 |
179 | **Key features:**
180 |
181 | * Expose MCP functionality to Goose through tools.
182 | * MCPs can be installed directly via the [extensions directory](https://block.github.io/goose/v1/extensions/), CLI, or UI.
183 | * Goose allows you to extend its functionality by [building your own MCP servers](https://block.github.io/goose/docs/tutorials/custom-extensions).
184 | * Includes built-in tools for development, web scraping, automation, memory, and integrations with JetBrains and Google Drive.
185 |
186 | ### LibreChat
187 |
188 | [LibreChat](https://github.com/danny-avila/LibreChat) is an open-source, customizable AI chat UI that supports multiple AI providers, now including MCP integration.
189 |
190 | **Key features:**
191 |
192 | * Extend current tool ecosystem, including [Code Interpreter](https://www.librechat.ai/docs/features/code_interpreter) and Image generation tools, through MCP servers
193 | * Add tools to customizable [Agents](https://www.librechat.ai/docs/features/agents), using a variety of LLMs from top providers
194 | * Open-source and self-hostable, with secure multi-user support
195 | * Future roadmap includes expanded MCP feature support
196 |
197 | ### mcp-agent
198 |
199 | [mcp-agent] is a simple, composable framework to build agents using Model Context Protocol.
200 |
201 | **Key features:**
202 |
203 | * Automatic connection management of MCP servers.
204 | * Expose tools from multiple servers to an LLM.
205 | * Implements every pattern defined in [Building Effective Agents](https://www.anthropic.com/research/building-effective-agents).
206 | * Supports workflow pause/resume signals, such as waiting for human feedback.
207 |
208 | ### Roo Code
209 |
210 | [Roo Code](https://roocode.com) enables AI coding assistance via MCP.
211 |
212 | **Key features:**
213 |
214 | * Support for MCP tools and resources
215 | * Integration with development workflows
216 | * Extensible AI capabilities
217 |
218 | ### Sourcegraph Cody
219 |
220 | [Cody](https://openctx.org/docs/providers/modelcontextprotocol) is Sourcegraph's AI coding assistant, which implements MCP through OpenCTX.
221 |
222 | **Key features:**
223 |
224 | * Support for MCP resources
225 | * Integration with Sourcegraph's code intelligence
226 | * Uses OpenCTX as an abstraction layer
227 | * Future support planned for additional MCP features
228 |
229 | ### Superinterface
230 |
231 | [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.
232 |
233 | **Key features:**
234 |
235 | * Use tools from MCP servers in assistants embedded via React components or script tags
236 | * SSE transport support
237 | * Use any AI model from any AI provider (OpenAI, Anthropic, Ollama, others)
238 |
239 | ### TheiaAI/TheiaIDE
240 |
241 | [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.
242 |
243 | **Key features:**
244 |
245 | * **Tool Integration**: Theia AI enables AI agents, including those in the Theia IDE, to utilize MCP servers for seamless tool interaction.
246 | * **Customizable Prompts**: The Theia IDE allows users to define and adapt prompts, dynamically integrating MCP servers for tailored workflows.
247 | * **Custom agents**: The Theia IDE supports creating custom agents that leverage MCP capabilities, enabling users to design dedicated workflows on the fly.
248 |
249 | Theia AI and Theia IDE's MCP integration provide users with flexibility, making them powerful platforms for exploring and adapting MCP.
250 |
251 | **Learn more:**
252 |
253 | * [Theia IDE and Theia AI MCP Announcement](https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/)
254 | * [Download the AI-powered Theia IDE](https://theia-ide.org/)
255 |
256 | ### Windsurf Editor
257 |
258 | [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.
259 |
260 | **Key features:**
261 |
262 | * Revolutionary AI Flow paradigm for human-AI collaboration
263 | * Intelligent code generation and understanding
264 | * Rich development tools with multi-model support
265 |
266 | ### Zed
267 |
268 | [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.
269 |
270 | **Key features:**
271 |
272 | * Prompt templates surface as slash commands in the editor
273 | * Tool integration for enhanced coding workflows
274 | * Tight integration with editor features and workspace context
275 | * Does not support MCP resources
276 |
277 | ### OpenSumi
278 |
279 | [OpenSumi](https://github.com/opensumi/core) is a framework helps you quickly build AI Native IDE products.
280 |
281 | **Key features:**
282 |
283 | * Supports MCP tools in OpenSumi
284 | * Supports built-in IDE MCP servers and custom MCP servers
285 |
286 | ## Adding MCP support to your application
287 |
288 | 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.
289 |
290 | Benefits of adding MCP support:
291 |
292 | * Enable users to bring their own context and tools
293 | * Join a growing ecosystem of interoperable AI applications
294 | * Provide users with flexible integration options
295 | * Support local-first AI workflows
296 |
297 | 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)
298 |
299 | ## Updates and corrections
300 |
301 | 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).
302 |
303 |
304 | # Contributing
305 | Source: https://modelcontextprotocol.io/development/contributing
306 |
307 | How to participate in Model Context Protocol development
308 |
309 | 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.
310 |
311 | All contributors must adhere to our [Code of Conduct](https://github.com/modelcontextprotocol/.github/blob/main/CODE_OF_CONDUCT.md).
312 |
313 | For questions and discussions, please use [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions).
314 |
315 |
316 | # Roadmap
317 | Source: https://modelcontextprotocol.io/development/roadmap
318 |
319 | Our plans for evolving Model Context Protocol (H1 2025)
320 |
321 | 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.
322 |
323 | <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>
324 |
325 | We encourage community participation! Each section links to relevant discussions where you can learn more and contribute your thoughts.
326 |
327 | ## Remote MCP Support
328 |
329 | 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:
330 |
331 | * [**Authentication & Authorization**](https://github.com/modelcontextprotocol/specification/discussions/64): Adding standardized auth capabilities, particularly focused on OAuth 2.0 support.
332 |
333 | * [**Service Discovery**](https://github.com/modelcontextprotocol/specification/discussions/69): Defining how clients can discover and connect to remote MCP servers.
334 |
335 | * [**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.
336 |
337 | ## Reference Implementations
338 |
339 | To help developers build with MCP, we want to offer documentation for:
340 |
341 | * **Client Examples**: Comprehensive reference client implementation(s), demonstrating all protocol features
342 | * **Protocol Drafting**: Streamlined process for proposing and incorporating new protocol features
343 |
344 | ## Distribution & Discovery
345 |
346 | Looking ahead, we're exploring ways to make MCP servers more accessible. Some areas we may investigate include:
347 |
348 | * **Package Management**: Standardized packaging format for MCP servers
349 | * **Installation Tools**: Simplified server installation across MCP clients
350 | * **Sandboxing**: Improved security through server isolation
351 | * **Server Registry**: A common directory for discovering available MCP servers
352 |
353 | ## Agent Support
354 |
355 | We're expanding MCP's capabilities for [complex agentic workflows](https://github.com/modelcontextprotocol/specification/discussions/111), particularly focusing on:
356 |
357 | * [**Hierarchical Agent Systems**](https://github.com/modelcontextprotocol/specification/discussions/94): Improved support for trees of agents through namespacing and topology awareness.
358 |
359 | * [**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.
360 |
361 | * [**Streaming Results**](https://github.com/modelcontextprotocol/specification/issues/117): Real-time updates from long-running agent operations.
362 |
363 | ## Broader Ecosystem
364 |
365 | We're also invested in:
366 |
367 | * **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.
368 | * [**Additional Modalities**](https://github.com/modelcontextprotocol/specification/discussions/88): Expanding beyond text to support audio, video, and other formats.
369 | * \[**Standardization**] Considering standardization through a standardization body.
370 |
371 | ## Get Involved
372 |
373 | 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.
374 |
375 |
376 | # What's New
377 | Source: https://modelcontextprotocol.io/development/updates
378 |
379 | The latest updates and improvements to MCP
380 |
381 | <Update label="2025-02-14" description="Java SDK released">
382 | * We're excited to announce that the Java SDK developed by Spring AI at VMware Tanzu is now
383 | the official [Java SDK](https://github.com/modelcontextprotocol/java-sdk) for MCP.
384 | This joins our existing Kotlin SDK in our growing list of supported languages.
385 | The Spring AI team will maintain the SDK as an integral part of the Model Context Protocol
386 | organization. We're thrilled to welcome them to the MCP community!
387 | </Update>
388 |
389 | <Update label="2025-01-27" description="Python SDK 1.2.1">
390 | * Version [1.2.1](https://github.com/modelcontextprotocol/python-sdk/releases/tag/v1.2.1) of the MCP Python SDK has been released,
391 | delivering important stability improvements and bug fixes.
392 | </Update>
393 |
394 | <Update label="2025-01-18" description="SDK and Server Improvements">
395 | * Simplified, express-like API in the [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)
396 | * Added 8 new clients to the [clients page](https://modelcontextprotocol.io/clients)
397 | </Update>
398 |
399 | <Update label="2025-01-03" description="SDK and Server Improvements">
400 | * FastMCP API in the [Python SDK](https://github.com/modelcontextprotocol/python-sdk)
401 | * Dockerized MCP servers in the [servers repo](https://github.com/modelcontextprotocol/servers)
402 | </Update>
403 |
404 | <Update label="2024-12-21" description="Kotlin SDK released">
405 | * Jetbrains released a Kotlin SDK for MCP!
406 | * For a sample MCP Kotlin server, check out [this repository](https://github.com/modelcontextprotocol/kotlin-sdk/tree/main/samples/kotlin-mcp-server)
407 | </Update>
408 |
409 |
410 | # Core architecture
411 | Source: https://modelcontextprotocol.io/docs/concepts/architecture
412 |
413 | Understand how MCP connects clients, servers, and LLMs
414 |
415 | 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.
416 |
417 | ## Overview
418 |
419 | MCP follows a client-server architecture where:
420 |
421 | * **Hosts** are LLM applications (like Claude Desktop or IDEs) that initiate connections
422 | * **Clients** maintain 1:1 connections with servers, inside the host application
423 | * **Servers** provide context, tools, and prompts to clients
424 |
425 | ```mermaid
426 | flowchart LR
427 | subgraph " Host (e.g., Claude Desktop) "
428 | client1[MCP Client]
429 | client2[MCP Client]
430 | end
431 | subgraph "Server Process"
432 | server1[MCP Server]
433 | end
434 | subgraph "Server Process"
435 | server2[MCP Server]
436 | end
437 |
438 | client1 <-->|Transport Layer| server1
439 | client2 <-->|Transport Layer| server2
440 | ```
441 |
442 | ## Core components
443 |
444 | ### Protocol layer
445 |
446 | The protocol layer handles message framing, request/response linking, and high-level communication patterns.
447 |
448 | <Tabs>
449 | <Tab title="TypeScript">
450 | ```typescript
451 | class Protocol<Request, Notification, Result> {
452 | // Handle incoming requests
453 | setRequestHandler<T>(schema: T, handler: (request: T, extra: RequestHandlerExtra) => Promise<Result>): void
454 |
455 | // Handle incoming notifications
456 | setNotificationHandler<T>(schema: T, handler: (notification: T) => Promise<void>): void
457 |
458 | // Send requests and await responses
459 | request<T>(request: Request, schema: T, options?: RequestOptions): Promise<T>
460 |
461 | // Send one-way notifications
462 | notification(notification: Notification): Promise<void>
463 | }
464 | ```
465 | </Tab>
466 |
467 | <Tab title="Python">
468 | ```python
469 | class Session(BaseSession[RequestT, NotificationT, ResultT]):
470 | async def send_request(
471 | self,
472 | request: RequestT,
473 | result_type: type[Result]
474 | ) -> Result:
475 | """
476 | Send request and wait for response. Raises McpError if response contains error.
477 | """
478 | # Request handling implementation
479 |
480 | async def send_notification(
481 | self,
482 | notification: NotificationT
483 | ) -> None:
484 | """Send one-way notification that doesn't expect response."""
485 | # Notification handling implementation
486 |
487 | async def _received_request(
488 | self,
489 | responder: RequestResponder[ReceiveRequestT, ResultT]
490 | ) -> None:
491 | """Handle incoming request from other side."""
492 | # Request handling implementation
493 |
494 | async def _received_notification(
495 | self,
496 | notification: ReceiveNotificationT
497 | ) -> None:
498 | """Handle incoming notification from other side."""
499 | # Notification handling implementation
500 | ```
501 | </Tab>
502 | </Tabs>
503 |
504 | Key classes include:
505 |
506 | * `Protocol`
507 | * `Client`
508 | * `Server`
509 |
510 | ### Transport layer
511 |
512 | The transport layer handles the actual communication between clients and servers. MCP supports multiple transport mechanisms:
513 |
514 | 1. **Stdio transport**
515 | * Uses standard input/output for communication
516 | * Ideal for local processes
517 |
518 | 2. **HTTP with SSE transport**
519 | * Uses Server-Sent Events for server-to-client messages
520 | * HTTP POST for client-to-server messages
521 |
522 | 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.
523 |
524 | ### Message types
525 |
526 | MCP has these main types of messages:
527 |
528 | 1. **Requests** expect a response from the other side:
529 | ```typescript
530 | interface Request {
531 | method: string;
532 | params?: { ... };
533 | }
534 | ```
535 |
536 | 2. **Results** are successful responses to requests:
537 | ```typescript
538 | interface Result {
539 | [key: string]: unknown;
540 | }
541 | ```
542 |
543 | 3. **Errors** indicate that a request failed:
544 | ```typescript
545 | interface Error {
546 | code: number;
547 | message: string;
548 | data?: unknown;
549 | }
550 | ```
551 |
552 | 4. **Notifications** are one-way messages that don't expect a response:
553 | ```typescript
554 | interface Notification {
555 | method: string;
556 | params?: { ... };
557 | }
558 | ```
559 |
560 | ## Connection lifecycle
561 |
562 | ### 1. Initialization
563 |
564 | ```mermaid
565 | sequenceDiagram
566 | participant Client
567 | participant Server
568 |
569 | Client->>Server: initialize request
570 | Server->>Client: initialize response
571 | Client->>Server: initialized notification
572 |
573 | Note over Client,Server: Connection ready for use
574 | ```
575 |
576 | 1. Client sends `initialize` request with protocol version and capabilities
577 | 2. Server responds with its protocol version and capabilities
578 | 3. Client sends `initialized` notification as acknowledgment
579 | 4. Normal message exchange begins
580 |
581 | ### 2. Message exchange
582 |
583 | After initialization, the following patterns are supported:
584 |
585 | * **Request-Response**: Client or server sends requests, the other responds
586 | * **Notifications**: Either party sends one-way messages
587 |
588 | ### 3. Termination
589 |
590 | Either party can terminate the connection:
591 |
592 | * Clean shutdown via `close()`
593 | * Transport disconnection
594 | * Error conditions
595 |
596 | ## Error handling
597 |
598 | MCP defines these standard error codes:
599 |
600 | ```typescript
601 | enum ErrorCode {
602 | // Standard JSON-RPC error codes
603 | ParseError = -32700,
604 | InvalidRequest = -32600,
605 | MethodNotFound = -32601,
606 | InvalidParams = -32602,
607 | InternalError = -32603
608 | }
609 | ```
610 |
611 | SDKs and applications can define their own error codes above -32000.
612 |
613 | Errors are propagated through:
614 |
615 | * Error responses to requests
616 | * Error events on transports
617 | * Protocol-level error handlers
618 |
619 | ## Implementation example
620 |
621 | Here's a basic example of implementing an MCP server:
622 |
623 | <Tabs>
624 | <Tab title="TypeScript">
625 | ```typescript
626 | import { Server } from "@modelcontextprotocol/sdk/server/index.js";
627 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
628 |
629 | const server = new Server({
630 | name: "example-server",
631 | version: "1.0.0"
632 | }, {
633 | capabilities: {
634 | resources: {}
635 | }
636 | });
637 |
638 | // Handle requests
639 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
640 | return {
641 | resources: [
642 | {
643 | uri: "example://resource",
644 | name: "Example Resource"
645 | }
646 | ]
647 | };
648 | });
649 |
650 | // Connect transport
651 | const transport = new StdioServerTransport();
652 | await server.connect(transport);
653 | ```
654 | </Tab>
655 |
656 | <Tab title="Python">
657 | ```python
658 | import asyncio
659 | import mcp.types as types
660 | from mcp.server import Server
661 | from mcp.server.stdio import stdio_server
662 |
663 | app = Server("example-server")
664 |
665 | @app.list_resources()
666 | async def list_resources() -> list[types.Resource]:
667 | return [
668 | types.Resource(
669 | uri="example://resource",
670 | name="Example Resource"
671 | )
672 | ]
673 |
674 | async def main():
675 | async with stdio_server() as streams:
676 | await app.run(
677 | streams[0],
678 | streams[1],
679 | app.create_initialization_options()
680 | )
681 |
682 | if __name__ == "__main__":
683 | asyncio.run(main)
684 | ```
685 | </Tab>
686 | </Tabs>
687 |
688 | ## Best practices
689 |
690 | ### Transport selection
691 |
692 | 1. **Local communication**
693 | * Use stdio transport for local processes
694 | * Efficient for same-machine communication
695 | * Simple process management
696 |
697 | 2. **Remote communication**
698 | * Use SSE for scenarios requiring HTTP compatibility
699 | * Consider security implications including authentication and authorization
700 |
701 | ### Message handling
702 |
703 | 1. **Request processing**
704 | * Validate inputs thoroughly
705 | * Use type-safe schemas
706 | * Handle errors gracefully
707 | * Implement timeouts
708 |
709 | 2. **Progress reporting**
710 | * Use progress tokens for long operations
711 | * Report progress incrementally
712 | * Include total progress when known
713 |
714 | 3. **Error management**
715 | * Use appropriate error codes
716 | * Include helpful error messages
717 | * Clean up resources on errors
718 |
719 | ## Security considerations
720 |
721 | 1. **Transport security**
722 | * Use TLS for remote connections
723 | * Validate connection origins
724 | * Implement authentication when needed
725 |
726 | 2. **Message validation**
727 | * Validate all incoming messages
728 | * Sanitize inputs
729 | * Check message size limits
730 | * Verify JSON-RPC format
731 |
732 | 3. **Resource protection**
733 | * Implement access controls
734 | * Validate resource paths
735 | * Monitor resource usage
736 | * Rate limit requests
737 |
738 | 4. **Error handling**
739 | * Don't leak sensitive information
740 | * Log security-relevant errors
741 | * Implement proper cleanup
742 | * Handle DoS scenarios
743 |
744 | ## Debugging and monitoring
745 |
746 | 1. **Logging**
747 | * Log protocol events
748 | * Track message flow
749 | * Monitor performance
750 | * Record errors
751 |
752 | 2. **Diagnostics**
753 | * Implement health checks
754 | * Monitor connection state
755 | * Track resource usage
756 | * Profile performance
757 |
758 | 3. **Testing**
759 | * Test different transports
760 | * Verify error handling
761 | * Check edge cases
762 | * Load test servers
763 |
764 |
765 | # Prompts
766 | Source: https://modelcontextprotocol.io/docs/concepts/prompts
767 |
768 | Create reusable prompt templates and workflows
769 |
770 | 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.
771 |
772 | <Note>
773 | 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.
774 | </Note>
775 |
776 | ## Overview
777 |
778 | Prompts in MCP are predefined templates that can:
779 |
780 | * Accept dynamic arguments
781 | * Include context from resources
782 | * Chain multiple interactions
783 | * Guide specific workflows
784 | * Surface as UI elements (like slash commands)
785 |
786 | ## Prompt structure
787 |
788 | Each prompt is defined with:
789 |
790 | ```typescript
791 | {
792 | name: string; // Unique identifier for the prompt
793 | description?: string; // Human-readable description
794 | arguments?: [ // Optional list of arguments
795 | {
796 | name: string; // Argument identifier
797 | description?: string; // Argument description
798 | required?: boolean; // Whether argument is required
799 | }
800 | ]
801 | }
802 | ```
803 |
804 | ## Discovering prompts
805 |
806 | Clients can discover available prompts through the `prompts/list` endpoint:
807 |
808 | ```typescript
809 | // Request
810 | {
811 | method: "prompts/list"
812 | }
813 |
814 | // Response
815 | {
816 | prompts: [
817 | {
818 | name: "analyze-code",
819 | description: "Analyze code for potential improvements",
820 | arguments: [
821 | {
822 | name: "language",
823 | description: "Programming language",
824 | required: true
825 | }
826 | ]
827 | }
828 | ]
829 | }
830 | ```
831 |
832 | ## Using prompts
833 |
834 | To use a prompt, clients make a `prompts/get` request:
835 |
836 | ````typescript
837 | // Request
838 | {
839 | method: "prompts/get",
840 | params: {
841 | name: "analyze-code",
842 | arguments: {
843 | language: "python"
844 | }
845 | }
846 | }
847 |
848 | // Response
849 | {
850 | description: "Analyze Python code for potential improvements",
851 | messages: [
852 | {
853 | role: "user",
854 | content: {
855 | type: "text",
856 | 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```"
857 | }
858 | }
859 | ]
860 | }
861 | ````
862 |
863 | ## Dynamic prompts
864 |
865 | Prompts can be dynamic and include:
866 |
867 | ### Embedded resource context
868 |
869 | ```json
870 | {
871 | "name": "analyze-project",
872 | "description": "Analyze project logs and code",
873 | "arguments": [
874 | {
875 | "name": "timeframe",
876 | "description": "Time period to analyze logs",
877 | "required": true
878 | },
879 | {
880 | "name": "fileUri",
881 | "description": "URI of code file to review",
882 | "required": true
883 | }
884 | ]
885 | }
886 | ```
887 |
888 | When handling the `prompts/get` request:
889 |
890 | ```json
891 | {
892 | "messages": [
893 | {
894 | "role": "user",
895 | "content": {
896 | "type": "text",
897 | "text": "Analyze these system logs and the code file for any issues:"
898 | }
899 | },
900 | {
901 | "role": "user",
902 | "content": {
903 | "type": "resource",
904 | "resource": {
905 | "uri": "logs://recent?timeframe=1h",
906 | "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",
907 | "mimeType": "text/plain"
908 | }
909 | }
910 | },
911 | {
912 | "role": "user",
913 | "content": {
914 | "type": "resource",
915 | "resource": {
916 | "uri": "file:///path/to/code.py",
917 | "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",
918 | "mimeType": "text/x-python"
919 | }
920 | }
921 | }
922 | ]
923 | }
924 | ```
925 |
926 | ### Multi-step workflows
927 |
928 | ```typescript
929 | const debugWorkflow = {
930 | name: "debug-error",
931 | async getMessages(error: string) {
932 | return [
933 | {
934 | role: "user",
935 | content: {
936 | type: "text",
937 | text: `Here's an error I'm seeing: ${error}`
938 | }
939 | },
940 | {
941 | role: "assistant",
942 | content: {
943 | type: "text",
944 | text: "I'll help analyze this error. What have you tried so far?"
945 | }
946 | },
947 | {
948 | role: "user",
949 | content: {
950 | type: "text",
951 | text: "I've tried restarting the service, but the error persists."
952 | }
953 | }
954 | ];
955 | }
956 | };
957 | ```
958 |
959 | ## Example implementation
960 |
961 | Here's a complete example of implementing prompts in an MCP server:
962 |
963 | <Tabs>
964 | <Tab title="TypeScript">
965 | ```typescript
966 | import { Server } from "@modelcontextprotocol/sdk/server";
967 | import {
968 | ListPromptsRequestSchema,
969 | GetPromptRequestSchema
970 | } from "@modelcontextprotocol/sdk/types";
971 |
972 | const PROMPTS = {
973 | "git-commit": {
974 | name: "git-commit",
975 | description: "Generate a Git commit message",
976 | arguments: [
977 | {
978 | name: "changes",
979 | description: "Git diff or description of changes",
980 | required: true
981 | }
982 | ]
983 | },
984 | "explain-code": {
985 | name: "explain-code",
986 | description: "Explain how code works",
987 | arguments: [
988 | {
989 | name: "code",
990 | description: "Code to explain",
991 | required: true
992 | },
993 | {
994 | name: "language",
995 | description: "Programming language",
996 | required: false
997 | }
998 | ]
999 | }
1000 | };
1001 |
1002 | const server = new Server({
1003 | name: "example-prompts-server",
1004 | version: "1.0.0"
1005 | }, {
1006 | capabilities: {
1007 | prompts: {}
1008 | }
1009 | });
1010 |
1011 | // List available prompts
1012 | server.setRequestHandler(ListPromptsRequestSchema, async () => {
1013 | return {
1014 | prompts: Object.values(PROMPTS)
1015 | };
1016 | });
1017 |
1018 | // Get specific prompt
1019 | server.setRequestHandler(GetPromptRequestSchema, async (request) => {
1020 | const prompt = PROMPTS[request.params.name];
1021 | if (!prompt) {
1022 | throw new Error(`Prompt not found: ${request.params.name}`);
1023 | }
1024 |
1025 | if (request.params.name === "git-commit") {
1026 | return {
1027 | messages: [
1028 | {
1029 | role: "user",
1030 | content: {
1031 | type: "text",
1032 | text: `Generate a concise but descriptive commit message for these changes:\n\n${request.params.arguments?.changes}`
1033 | }
1034 | }
1035 | ]
1036 | };
1037 | }
1038 |
1039 | if (request.params.name === "explain-code") {
1040 | const language = request.params.arguments?.language || "Unknown";
1041 | return {
1042 | messages: [
1043 | {
1044 | role: "user",
1045 | content: {
1046 | type: "text",
1047 | text: `Explain how this ${language} code works:\n\n${request.params.arguments?.code}`
1048 | }
1049 | }
1050 | ]
1051 | };
1052 | }
1053 |
1054 | throw new Error("Prompt implementation not found");
1055 | });
1056 | ```
1057 | </Tab>
1058 |
1059 | <Tab title="Python">
1060 | ```python
1061 | from mcp.server import Server
1062 | import mcp.types as types
1063 |
1064 | # Define available prompts
1065 | PROMPTS = {
1066 | "git-commit": types.Prompt(
1067 | name="git-commit",
1068 | description="Generate a Git commit message",
1069 | arguments=[
1070 | types.PromptArgument(
1071 | name="changes",
1072 | description="Git diff or description of changes",
1073 | required=True
1074 | )
1075 | ],
1076 | ),
1077 | "explain-code": types.Prompt(
1078 | name="explain-code",
1079 | description="Explain how code works",
1080 | arguments=[
1081 | types.PromptArgument(
1082 | name="code",
1083 | description="Code to explain",
1084 | required=True
1085 | ),
1086 | types.PromptArgument(
1087 | name="language",
1088 | description="Programming language",
1089 | required=False
1090 | )
1091 | ],
1092 | )
1093 | }
1094 |
1095 | # Initialize server
1096 | app = Server("example-prompts-server")
1097 |
1098 | @app.list_prompts()
1099 | async def list_prompts() -> list[types.Prompt]:
1100 | return list(PROMPTS.values())
1101 |
1102 | @app.get_prompt()
1103 | async def get_prompt(
1104 | name: str, arguments: dict[str, str] | None = None
1105 | ) -> types.GetPromptResult:
1106 | if name not in PROMPTS:
1107 | raise ValueError(f"Prompt not found: {name}")
1108 |
1109 | if name == "git-commit":
1110 | changes = arguments.get("changes") if arguments else ""
1111 | return types.GetPromptResult(
1112 | messages=[
1113 | types.PromptMessage(
1114 | role="user",
1115 | content=types.TextContent(
1116 | type="text",
1117 | text=f"Generate a concise but descriptive commit message "
1118 | f"for these changes:\n\n{changes}"
1119 | )
1120 | )
1121 | ]
1122 | )
1123 |
1124 | if name == "explain-code":
1125 | code = arguments.get("code") if arguments else ""
1126 | language = arguments.get("language", "Unknown") if arguments else "Unknown"
1127 | return types.GetPromptResult(
1128 | messages=[
1129 | types.PromptMessage(
1130 | role="user",
1131 | content=types.TextContent(
1132 | type="text",
1133 | text=f"Explain how this {language} code works:\n\n{code}"
1134 | )
1135 | )
1136 | ]
1137 | )
1138 |
1139 | raise ValueError("Prompt implementation not found")
1140 | ```
1141 | </Tab>
1142 | </Tabs>
1143 |
1144 | ## Best practices
1145 |
1146 | When implementing prompts:
1147 |
1148 | 1. Use clear, descriptive prompt names
1149 | 2. Provide detailed descriptions for prompts and arguments
1150 | 3. Validate all required arguments
1151 | 4. Handle missing arguments gracefully
1152 | 5. Consider versioning for prompt templates
1153 | 6. Cache dynamic content when appropriate
1154 | 7. Implement error handling
1155 | 8. Document expected argument formats
1156 | 9. Consider prompt composability
1157 | 10. Test prompts with various inputs
1158 |
1159 | ## UI integration
1160 |
1161 | Prompts can be surfaced in client UIs as:
1162 |
1163 | * Slash commands
1164 | * Quick actions
1165 | * Context menu items
1166 | * Command palette entries
1167 | * Guided workflows
1168 | * Interactive forms
1169 |
1170 | ## Updates and changes
1171 |
1172 | Servers can notify clients about prompt changes:
1173 |
1174 | 1. Server capability: `prompts.listChanged`
1175 | 2. Notification: `notifications/prompts/list_changed`
1176 | 3. Client re-fetches prompt list
1177 |
1178 | ## Security considerations
1179 |
1180 | When implementing prompts:
1181 |
1182 | * Validate all arguments
1183 | * Sanitize user input
1184 | * Consider rate limiting
1185 | * Implement access controls
1186 | * Audit prompt usage
1187 | * Handle sensitive data appropriately
1188 | * Validate generated content
1189 | * Implement timeouts
1190 | * Consider prompt injection risks
1191 | * Document security requirements
1192 |
1193 |
1194 | # Resources
1195 | Source: https://modelcontextprotocol.io/docs/concepts/resources
1196 |
1197 | Expose data and content from your servers to LLMs
1198 |
1199 | 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.
1200 |
1201 | <Note>
1202 | Resources are designed to be **application-controlled**, meaning that the client application can decide how and when they should be used.
1203 | Different MCP clients may handle resources differently. For example:
1204 |
1205 | * Claude Desktop currently requires users to explicitly select resources before they can be used
1206 | * Other clients might automatically select resources based on heuristics
1207 | * Some implementations may even allow the AI model itself to determine which resources to use
1208 |
1209 | 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).
1210 | </Note>
1211 |
1212 | ## Overview
1213 |
1214 | Resources represent any kind of data that an MCP server wants to make available to clients. This can include:
1215 |
1216 | * File contents
1217 | * Database records
1218 | * API responses
1219 | * Live system data
1220 | * Screenshots and images
1221 | * Log files
1222 | * And more
1223 |
1224 | Each resource is identified by a unique URI and can contain either text or binary data.
1225 |
1226 | ## Resource URIs
1227 |
1228 | Resources are identified using URIs that follow this format:
1229 |
1230 | ```
1231 | [protocol]://[host]/[path]
1232 | ```
1233 |
1234 | For example:
1235 |
1236 | * `file:///home/user/documents/report.pdf`
1237 | * `postgres://database/customers/schema`
1238 | * `screen://localhost/display1`
1239 |
1240 | The protocol and path structure is defined by the MCP server implementation. Servers can define their own custom URI schemes.
1241 |
1242 | ## Resource types
1243 |
1244 | Resources can contain two types of content:
1245 |
1246 | ### Text resources
1247 |
1248 | Text resources contain UTF-8 encoded text data. These are suitable for:
1249 |
1250 | * Source code
1251 | * Configuration files
1252 | * Log files
1253 | * JSON/XML data
1254 | * Plain text
1255 |
1256 | ### Binary resources
1257 |
1258 | Binary resources contain raw binary data encoded in base64. These are suitable for:
1259 |
1260 | * Images
1261 | * PDFs
1262 | * Audio files
1263 | * Video files
1264 | * Other non-text formats
1265 |
1266 | ## Resource discovery
1267 |
1268 | Clients can discover available resources through two main methods:
1269 |
1270 | ### Direct resources
1271 |
1272 | Servers expose a list of concrete resources via the `resources/list` endpoint. Each resource includes:
1273 |
1274 | ```typescript
1275 | {
1276 | uri: string; // Unique identifier for the resource
1277 | name: string; // Human-readable name
1278 | description?: string; // Optional description
1279 | mimeType?: string; // Optional MIME type
1280 | }
1281 | ```
1282 |
1283 | ### Resource templates
1284 |
1285 | For dynamic resources, servers can expose [URI templates](https://datatracker.ietf.org/doc/html/rfc6570) that clients can use to construct valid resource URIs:
1286 |
1287 | ```typescript
1288 | {
1289 | uriTemplate: string; // URI template following RFC 6570
1290 | name: string; // Human-readable name for this type
1291 | description?: string; // Optional description
1292 | mimeType?: string; // Optional MIME type for all matching resources
1293 | }
1294 | ```
1295 |
1296 | ## Reading resources
1297 |
1298 | To read a resource, clients make a `resources/read` request with the resource URI.
1299 |
1300 | The server responds with a list of resource contents:
1301 |
1302 | ```typescript
1303 | {
1304 | contents: [
1305 | {
1306 | uri: string; // The URI of the resource
1307 | mimeType?: string; // Optional MIME type
1308 |
1309 | // One of:
1310 | text?: string; // For text resources
1311 | blob?: string; // For binary resources (base64 encoded)
1312 | }
1313 | ]
1314 | }
1315 | ```
1316 |
1317 | <Tip>
1318 | 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.
1319 | </Tip>
1320 |
1321 | ## Resource updates
1322 |
1323 | MCP supports real-time updates for resources through two mechanisms:
1324 |
1325 | ### List changes
1326 |
1327 | Servers can notify clients when their list of available resources changes via the `notifications/resources/list_changed` notification.
1328 |
1329 | ### Content changes
1330 |
1331 | Clients can subscribe to updates for specific resources:
1332 |
1333 | 1. Client sends `resources/subscribe` with resource URI
1334 | 2. Server sends `notifications/resources/updated` when the resource changes
1335 | 3. Client can fetch latest content with `resources/read`
1336 | 4. Client can unsubscribe with `resources/unsubscribe`
1337 |
1338 | ## Example implementation
1339 |
1340 | Here's a simple example of implementing resource support in an MCP server:
1341 |
1342 | <Tabs>
1343 | <Tab title="TypeScript">
1344 | ```typescript
1345 | const server = new Server({
1346 | name: "example-server",
1347 | version: "1.0.0"
1348 | }, {
1349 | capabilities: {
1350 | resources: {}
1351 | }
1352 | });
1353 |
1354 | // List available resources
1355 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
1356 | return {
1357 | resources: [
1358 | {
1359 | uri: "file:///logs/app.log",
1360 | name: "Application Logs",
1361 | mimeType: "text/plain"
1362 | }
1363 | ]
1364 | };
1365 | });
1366 |
1367 | // Read resource contents
1368 | server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
1369 | const uri = request.params.uri;
1370 |
1371 | if (uri === "file:///logs/app.log") {
1372 | const logContents = await readLogFile();
1373 | return {
1374 | contents: [
1375 | {
1376 | uri,
1377 | mimeType: "text/plain",
1378 | text: logContents
1379 | }
1380 | ]
1381 | };
1382 | }
1383 |
1384 | throw new Error("Resource not found");
1385 | });
1386 | ```
1387 | </Tab>
1388 |
1389 | <Tab title="Python">
1390 | ```python
1391 | app = Server("example-server")
1392 |
1393 | @app.list_resources()
1394 | async def list_resources() -> list[types.Resource]:
1395 | return [
1396 | types.Resource(
1397 | uri="file:///logs/app.log",
1398 | name="Application Logs",
1399 | mimeType="text/plain"
1400 | )
1401 | ]
1402 |
1403 | @app.read_resource()
1404 | async def read_resource(uri: AnyUrl) -> str:
1405 | if str(uri) == "file:///logs/app.log":
1406 | log_contents = await read_log_file()
1407 | return log_contents
1408 |
1409 | raise ValueError("Resource not found")
1410 |
1411 | # Start server
1412 | async with stdio_server() as streams:
1413 | await app.run(
1414 | streams[0],
1415 | streams[1],
1416 | app.create_initialization_options()
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 |
1453 | # Roots
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 |
1534 | # Sampling
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 | * `name`: String that can match full or partial model names (e.g. "claude-3", "sonnet")
1612 | * Clients may map hints to equivalent models from different providers
1613 | * Multiple hints are evaluated in preference order
1614 |
1615 | * Priority values (0-1 normalized):
1616 | * `costPriority`: Importance of minimizing costs
1617 | * `speedPriority`: Importance of low latency response
1618 | * `intelligencePriority`: Importance of advanced model capabilities
1619 |
1620 | Clients make the final model selection based on these preferences and their available models.
1621 |
1622 | ### System prompt
1623 |
1624 | An optional `systemPrompt` field allows servers to request a specific system prompt. The client may modify or ignore this.
1625 |
1626 | ### Context inclusion
1627 |
1628 | The `includeContext` parameter specifies what MCP context to include:
1629 |
1630 | * `"none"`: No additional context
1631 | * `"thisServer"`: Include context from the requesting server
1632 | * `"allServers"`: Include context from all connected MCP servers
1633 |
1634 | The client controls what context is actually included.
1635 |
1636 | ### Sampling parameters
1637 |
1638 | Fine-tune the LLM sampling with:
1639 |
1640 | * `temperature`: Controls randomness (0.0 to 1.0)
1641 | * `maxTokens`: Maximum tokens to generate
1642 | * `stopSequences`: Array of sequences that stop generation
1643 | * `metadata`: Additional provider-specific parameters
1644 |
1645 | ## Response format
1646 |
1647 | The client returns a completion result:
1648 |
1649 | ```typescript
1650 | {
1651 | model: string, // Name of the model used
1652 | stopReason?: "endTurn" | "stopSequence" | "maxTokens" | string,
1653 | role: "user" | "assistant",
1654 | content: {
1655 | type: "text" | "image",
1656 | text?: string,
1657 | data?: string,
1658 | mimeType?: string
1659 | }
1660 | }
1661 | ```
1662 |
1663 | ## Example request
1664 |
1665 | Here's an example of requesting sampling from a client:
1666 |
1667 | ```json
1668 | {
1669 | "method": "sampling/createMessage",
1670 | "params": {
1671 | "messages": [
1672 | {
1673 | "role": "user",
1674 | "content": {
1675 | "type": "text",
1676 | "text": "What files are in the current directory?"
1677 | }
1678 | }
1679 | ],
1680 | "systemPrompt": "You are a helpful file system assistant.",
1681 | "includeContext": "thisServer",
1682 | "maxTokens": 100
1683 | }
1684 | }
1685 | ```
1686 |
1687 | ## Best practices
1688 |
1689 | When implementing sampling:
1690 |
1691 | 1. Always provide clear, well-structured prompts
1692 | 2. Handle both text and image content appropriately
1693 | 3. Set reasonable token limits
1694 | 4. Include relevant context through `includeContext`
1695 | 5. Validate responses before using them
1696 | 6. Handle errors gracefully
1697 | 7. Consider rate limiting sampling requests
1698 | 8. Document expected sampling behavior
1699 | 9. Test with various model parameters
1700 | 10. Monitor sampling costs
1701 |
1702 | ## Human in the loop controls
1703 |
1704 | Sampling is designed with human oversight in mind:
1705 |
1706 | ### For prompts
1707 |
1708 | * Clients should show users the proposed prompt
1709 | * Users should be able to modify or reject prompts
1710 | * System prompts can be filtered or modified
1711 | * Context inclusion is controlled by the client
1712 |
1713 | ### For completions
1714 |
1715 | * Clients should show users the completion
1716 | * Users should be able to modify or reject completions
1717 | * Clients can filter or modify completions
1718 | * Users control which model is used
1719 |
1720 | ## Security considerations
1721 |
1722 | When implementing sampling:
1723 |
1724 | * Validate all message content
1725 | * Sanitize sensitive information
1726 | * Implement appropriate rate limits
1727 | * Monitor sampling usage
1728 | * Encrypt data in transit
1729 | * Handle user data privacy
1730 | * Audit sampling requests
1731 | * Control cost exposure
1732 | * Implement timeouts
1733 | * Handle model errors gracefully
1734 |
1735 | ## Common patterns
1736 |
1737 | ### Agentic workflows
1738 |
1739 | Sampling enables agentic patterns like:
1740 |
1741 | * Reading and analyzing resources
1742 | * Making decisions based on context
1743 | * Generating structured data
1744 | * Handling multi-step tasks
1745 | * Providing interactive assistance
1746 |
1747 | ### Context management
1748 |
1749 | Best practices for context:
1750 |
1751 | * Request minimal necessary context
1752 | * Structure context clearly
1753 | * Handle context size limits
1754 | * Update context as needed
1755 | * Clean up stale context
1756 |
1757 | ### Error handling
1758 |
1759 | Robust error handling should:
1760 |
1761 | * Catch sampling failures
1762 | * Handle timeout errors
1763 | * Manage rate limits
1764 | * Validate responses
1765 | * Provide fallback behaviors
1766 | * Log errors appropriately
1767 |
1768 | ## Limitations
1769 |
1770 | Be aware of these limitations:
1771 |
1772 | * Sampling depends on client capabilities
1773 | * Users control sampling behavior
1774 | * Context size has limits
1775 | * Rate limits may apply
1776 | * Costs should be considered
1777 | * Model availability varies
1778 | * Response times vary
1779 | * Not all content types supported
1780 |
1781 |
1782 | # Tools
1783 | Source: https://modelcontextprotocol.io/docs/concepts/tools
1784 |
1785 | Enable LLMs to perform actions through your server
1786 |
1787 | 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.
1788 |
1789 | <Note>
1790 | 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).
1791 | </Note>
1792 |
1793 | ## Overview
1794 |
1795 | 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:
1796 |
1797 | * **Discovery**: Clients can list available tools through the `tools/list` endpoint
1798 | * **Invocation**: Tools are called using the `tools/call` endpoint, where servers perform the requested operation and return results
1799 | * **Flexibility**: Tools can range from simple calculations to complex API interactions
1800 |
1801 | 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.
1802 |
1803 | ## Tool definition structure
1804 |
1805 | Each tool is defined with the following structure:
1806 |
1807 | ```typescript
1808 | {
1809 | name: string; // Unique identifier for the tool
1810 | description?: string; // Human-readable description
1811 | inputSchema: { // JSON Schema for the tool's parameters
1812 | type: "object",
1813 | properties: { ... } // Tool-specific parameters
1814 | }
1815 | }
1816 | ```
1817 |
1818 | ## Implementing tools
1819 |
1820 | Here's an example of implementing a basic tool in an MCP server:
1821 |
1822 | <Tabs>
1823 | <Tab title="TypeScript">
1824 | ```typescript
1825 | const server = new Server({
1826 | name: "example-server",
1827 | version: "1.0.0"
1828 | }, {
1829 | capabilities: {
1830 | tools: {}
1831 | }
1832 | });
1833 |
1834 | // Define available tools
1835 | server.setRequestHandler(ListToolsRequestSchema, async () => {
1836 | return {
1837 | tools: [{
1838 | name: "calculate_sum",
1839 | description: "Add two numbers together",
1840 | inputSchema: {
1841 | type: "object",
1842 | properties: {
1843 | a: { type: "number" },
1844 | b: { type: "number" }
1845 | },
1846 | required: ["a", "b"]
1847 | }
1848 | }]
1849 | };
1850 | });
1851 |
1852 | // Handle tool execution
1853 | server.setRequestHandler(CallToolRequestSchema, async (request) => {
1854 | if (request.params.name === "calculate_sum") {
1855 | const { a, b } = request.params.arguments;
1856 | return {
1857 | content: [
1858 | {
1859 | type: "text",
1860 | text: String(a + b)
1861 | }
1862 | ]
1863 | };
1864 | }
1865 | throw new Error("Tool not found");
1866 | });
1867 | ```
1868 | </Tab>
1869 |
1870 | <Tab title="Python">
1871 | ```python
1872 | app = Server("example-server")
1873 |
1874 | @app.list_tools()
1875 | async def list_tools() -> list[types.Tool]:
1876 | return [
1877 | types.Tool(
1878 | name="calculate_sum",
1879 | description="Add two numbers together",
1880 | inputSchema={
1881 | "type": "object",
1882 | "properties": {
1883 | "a": {"type": "number"},
1884 | "b": {"type": "number"}
1885 | },
1886 | "required": ["a", "b"]
1887 | }
1888 | )
1889 | ]
1890 |
1891 | @app.call_tool()
1892 | async def call_tool(
1893 | name: str,
1894 | arguments: dict
1895 | ) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
1896 | if name == "calculate_sum":
1897 | a = arguments["a"]
1898 | b = arguments["b"]
1899 | result = a + b
1900 | return [types.TextContent(type="text", text=str(result))]
1901 | raise ValueError(f"Tool not found: {name}")
1902 | ```
1903 | </Tab>
1904 | </Tabs>
1905 |
1906 | ## Example tool patterns
1907 |
1908 | Here are some examples of types of tools that a server could provide:
1909 |
1910 | ### System operations
1911 |
1912 | Tools that interact with the local system:
1913 |
1914 | ```typescript
1915 | {
1916 | name: "execute_command",
1917 | description: "Run a shell command",
1918 | inputSchema: {
1919 | type: "object",
1920 | properties: {
1921 | command: { type: "string" },
1922 | args: { type: "array", items: { type: "string" } }
1923 | }
1924 | }
1925 | }
1926 | ```
1927 |
1928 | ### API integrations
1929 |
1930 | Tools that wrap external APIs:
1931 |
1932 | ```typescript
1933 | {
1934 | name: "github_create_issue",
1935 | description: "Create a GitHub issue",
1936 | inputSchema: {
1937 | type: "object",
1938 | properties: {
1939 | title: { type: "string" },
1940 | body: { type: "string" },
1941 | labels: { type: "array", items: { type: "string" } }
1942 | }
1943 | }
1944 | }
1945 | ```
1946 |
1947 | ### Data processing
1948 |
1949 | Tools that transform or analyze data:
1950 |
1951 | ```typescript
1952 | {
1953 | name: "analyze_csv",
1954 | description: "Analyze a CSV file",
1955 | inputSchema: {
1956 | type: "object",
1957 | properties: {
1958 | filepath: { type: "string" },
1959 | operations: {
1960 | type: "array",
1961 | items: {
1962 | enum: ["sum", "average", "count"]
1963 | }
1964 | }
1965 | }
1966 | }
1967 | }
1968 | ```
1969 |
1970 | ## Best practices
1971 |
1972 | When implementing tools:
1973 |
1974 | 1. Provide clear, descriptive names and descriptions
1975 | 2. Use detailed JSON Schema definitions for parameters
1976 | 3. Include examples in tool descriptions to demonstrate how the model should use them
1977 | 4. Implement proper error handling and validation
1978 | 5. Use progress reporting for long operations
1979 | 6. Keep tool operations focused and atomic
1980 | 7. Document expected return value structures
1981 | 8. Implement proper timeouts
1982 | 9. Consider rate limiting for resource-intensive operations
1983 | 10. Log tool usage for debugging and monitoring
1984 |
1985 | ## Security considerations
1986 |
1987 | When exposing tools:
1988 |
1989 | ### Input validation
1990 |
1991 | * Validate all parameters against the schema
1992 | * Sanitize file paths and system commands
1993 | * Validate URLs and external identifiers
1994 | * Check parameter sizes and ranges
1995 | * Prevent command injection
1996 |
1997 | ### Access control
1998 |
1999 | * Implement authentication where needed
2000 | * Use appropriate authorization checks
2001 | * Audit tool usage
2002 | * Rate limit requests
2003 | * Monitor for abuse
2004 |
2005 | ### Error handling
2006 |
2007 | * Don't expose internal errors to clients
2008 | * Log security-relevant errors
2009 | * Handle timeouts appropriately
2010 | * Clean up resources after errors
2011 | * Validate return values
2012 |
2013 | ## Tool discovery and updates
2014 |
2015 | MCP supports dynamic tool discovery:
2016 |
2017 | 1. Clients can list available tools at any time
2018 | 2. Servers can notify clients when tools change using `notifications/tools/list_changed`
2019 | 3. Tools can be added or removed during runtime
2020 | 4. Tool definitions can be updated (though this should be done carefully)
2021 |
2022 | ## Error handling
2023 |
2024 | 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:
2025 |
2026 | 1. Set `isError` to `true` in the result
2027 | 2. Include error details in the `content` array
2028 |
2029 | Here's an example of proper error handling for tools:
2030 |
2031 | <Tabs>
2032 | <Tab title="TypeScript">
2033 | ```typescript
2034 | try {
2035 | // Tool operation
2036 | const result = performOperation();
2037 | return {
2038 | content: [
2039 | {
2040 | type: "text",
2041 | text: `Operation successful: ${result}`
2042 | }
2043 | ]
2044 | };
2045 | } catch (error) {
2046 | return {
2047 | isError: true,
2048 | content: [
2049 | {
2050 | type: "text",
2051 | text: `Error: ${error.message}`
2052 | }
2053 | ]
2054 | };
2055 | }
2056 | ```
2057 | </Tab>
2058 |
2059 | <Tab title="Python">
2060 | ```python
2061 | try:
2062 | # Tool operation
2063 | result = perform_operation()
2064 | return types.CallToolResult(
2065 | content=[
2066 | types.TextContent(
2067 | type="text",
2068 | text=f"Operation successful: {result}"
2069 | )
2070 | ]
2071 | )
2072 | except Exception as error:
2073 | return types.CallToolResult(
2074 | isError=True,
2075 | content=[
2076 | types.TextContent(
2077 | type="text",
2078 | text=f"Error: {str(error)}"
2079 | )
2080 | ]
2081 | )
2082 | ```
2083 | </Tab>
2084 | </Tabs>
2085 |
2086 | This approach allows the LLM to see that an error occurred and potentially take corrective action or request human intervention.
2087 |
2088 | ## Testing tools
2089 |
2090 | A comprehensive testing strategy for MCP tools should cover:
2091 |
2092 | * **Functional testing**: Verify tools execute correctly with valid inputs and handle invalid inputs appropriately
2093 | * **Integration testing**: Test tool interaction with external systems using both real and mocked dependencies
2094 | * **Security testing**: Validate authentication, authorization, input sanitization, and rate limiting
2095 | * **Performance testing**: Check behavior under load, timeout handling, and resource cleanup
2096 | * **Error handling**: Ensure tools properly report errors through the MCP protocol and clean up resources
2097 |
2098 |
2099 | # Transports
2100 | Source: https://modelcontextprotocol.io/docs/concepts/transports
2101 |
2102 | Learn about MCP's communication mechanisms
2103 |
2104 | 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.
2105 |
2106 | ## Message Format
2107 |
2108 | 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.
2109 |
2110 | There are three types of JSON-RPC messages used:
2111 |
2112 | ### Requests
2113 |
2114 | ```typescript
2115 | {
2116 | jsonrpc: "2.0",
2117 | id: number | string,
2118 | method: string,
2119 | params?: object
2120 | }
2121 | ```
2122 |
2123 | ### Responses
2124 |
2125 | ```typescript
2126 | {
2127 | jsonrpc: "2.0",
2128 | id: number | string,
2129 | result?: object,
2130 | error?: {
2131 | code: number,
2132 | message: string,
2133 | data?: unknown
2134 | }
2135 | }
2136 | ```
2137 |
2138 | ### Notifications
2139 |
2140 | ```typescript
2141 | {
2142 | jsonrpc: "2.0",
2143 | method: string,
2144 | params?: object
2145 | }
2146 | ```
2147 |
2148 | ## Built-in Transport Types
2149 |
2150 | MCP includes two standard transport implementations:
2151 |
2152 | ### Standard Input/Output (stdio)
2153 |
2154 | The stdio transport enables communication through standard input and output streams. This is particularly useful for local integrations and command-line tools.
2155 |
2156 | Use stdio when:
2157 |
2158 | * Building command-line tools
2159 | * Implementing local integrations
2160 | * Needing simple process communication
2161 | * Working with shell scripts
2162 |
2163 | <Tabs>
2164 | <Tab title="TypeScript (Server)">
2165 | ```typescript
2166 | const server = new Server({
2167 | name: "example-server",
2168 | version: "1.0.0"
2169 | }, {
2170 | capabilities: {}
2171 | });
2172 |
2173 | const transport = new StdioServerTransport();
2174 | await server.connect(transport);
2175 | ```
2176 | </Tab>
2177 |
2178 | <Tab title="TypeScript (Client)">
2179 | ```typescript
2180 | const client = new Client({
2181 | name: "example-client",
2182 | version: "1.0.0"
2183 | }, {
2184 | capabilities: {}
2185 | });
2186 |
2187 | const transport = new StdioClientTransport({
2188 | command: "./server",
2189 | args: ["--option", "value"]
2190 | });
2191 | await client.connect(transport);
2192 | ```
2193 | </Tab>
2194 |
2195 | <Tab title="Python (Server)">
2196 | ```python
2197 | app = Server("example-server")
2198 |
2199 | async with stdio_server() as streams:
2200 | await app.run(
2201 | streams[0],
2202 | streams[1],
2203 | app.create_initialization_options()
2204 | )
2205 | ```
2206 | </Tab>
2207 |
2208 | <Tab title="Python (Client)">
2209 | ```python
2210 | params = StdioServerParameters(
2211 | command="./server",
2212 | args=["--option", "value"]
2213 | )
2214 |
2215 | async with stdio_client(params) as streams:
2216 | async with ClientSession(streams[0], streams[1]) as session:
2217 | await session.initialize()
2218 | ```
2219 | </Tab>
2220 | </Tabs>
2221 |
2222 | ### Server-Sent Events (SSE)
2223 |
2224 | SSE transport enables server-to-client streaming with HTTP POST requests for client-to-server communication.
2225 |
2226 | Use SSE when:
2227 |
2228 | * Only server-to-client streaming is needed
2229 | * Working with restricted networks
2230 | * Implementing simple updates
2231 |
2232 | <Tabs>
2233 | <Tab title="TypeScript (Server)">
2234 | ```typescript
2235 | import express from "express";
2236 |
2237 | const app = express();
2238 |
2239 | const server = new Server({
2240 | name: "example-server",
2241 | version: "1.0.0"
2242 | }, {
2243 | capabilities: {}
2244 | });
2245 |
2246 | let transport: SSEServerTransport | null = null;
2247 |
2248 | app.get("/sse", (req, res) => {
2249 | transport = new SSEServerTransport("/messages", res);
2250 | server.connect(transport);
2251 | });
2252 |
2253 | app.post("/messages", (req, res) => {
2254 | if (transport) {
2255 | transport.handlePostMessage(req, res);
2256 | }
2257 | });
2258 |
2259 | app.listen(3000);
2260 | ```
2261 | </Tab>
2262 |
2263 | <Tab title="TypeScript (Client)">
2264 | ```typescript
2265 | const client = new Client({
2266 | name: "example-client",
2267 | version: "1.0.0"
2268 | }, {
2269 | capabilities: {}
2270 | });
2271 |
2272 | const transport = new SSEClientTransport(
2273 | new URL("http://localhost:3000/sse")
2274 | );
2275 | await client.connect(transport);
2276 | ```
2277 | </Tab>
2278 |
2279 | <Tab title="Python (Server)">
2280 | ```python
2281 | from mcp.server.sse import SseServerTransport
2282 | from starlette.applications import Starlette
2283 | from starlette.routing import Route
2284 |
2285 | app = Server("example-server")
2286 | sse = SseServerTransport("/messages")
2287 |
2288 | async def handle_sse(scope, receive, send):
2289 | async with sse.connect_sse(scope, receive, send) as streams:
2290 | await app.run(streams[0], streams[1], app.create_initialization_options())
2291 |
2292 | async def handle_messages(scope, receive, send):
2293 | await sse.handle_post_message(scope, receive, send)
2294 |
2295 | starlette_app = Starlette(
2296 | routes=[
2297 | Route("/sse", endpoint=handle_sse),
2298 | Route("/messages", endpoint=handle_messages, methods=["POST"]),
2299 | ]
2300 | )
2301 | ```
2302 | </Tab>
2303 |
2304 | <Tab title="Python (Client)">
2305 | ```python
2306 | async with sse_client("http://localhost:8000/sse") as streams:
2307 | async with ClientSession(streams[0], streams[1]) as session:
2308 | await session.initialize()
2309 | ```
2310 | </Tab>
2311 | </Tabs>
2312 |
2313 | ## Custom Transports
2314 |
2315 | MCP makes it easy to implement custom transports for specific needs. Any transport implementation just needs to conform to the Transport interface:
2316 |
2317 | You can implement custom transports for:
2318 |
2319 | * Custom network protocols
2320 | * Specialized communication channels
2321 | * Integration with existing systems
2322 | * Performance optimization
2323 |
2324 | <Tabs>
2325 | <Tab title="TypeScript">
2326 | ```typescript
2327 | interface Transport {
2328 | // Start processing messages
2329 | start(): Promise<void>;
2330 |
2331 | // Send a JSON-RPC message
2332 | send(message: JSONRPCMessage): Promise<void>;
2333 |
2334 | // Close the connection
2335 | close(): Promise<void>;
2336 |
2337 | // Callbacks
2338 | onclose?: () => void;
2339 | onerror?: (error: Error) => void;
2340 | onmessage?: (message: JSONRPCMessage) => void;
2341 | }
2342 | ```
2343 | </Tab>
2344 |
2345 | <Tab title="Python">
2346 | Note that while MCP Servers are often implemented with asyncio, we recommend
2347 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2348 |
2349 | ```python
2350 | @contextmanager
2351 | async def create_transport(
2352 | read_stream: MemoryObjectReceiveStream[JSONRPCMessage | Exception],
2353 | write_stream: MemoryObjectSendStream[JSONRPCMessage]
2354 | ):
2355 | """
2356 | Transport interface for MCP.
2357 |
2358 | Args:
2359 | read_stream: Stream to read incoming messages from
2360 | write_stream: Stream to write outgoing messages to
2361 | """
2362 | async with anyio.create_task_group() as tg:
2363 | try:
2364 | # Start processing messages
2365 | tg.start_soon(lambda: process_messages(read_stream))
2366 |
2367 | # Send messages
2368 | async with write_stream:
2369 | yield write_stream
2370 |
2371 | except Exception as exc:
2372 | # Handle errors
2373 | raise exc
2374 | finally:
2375 | # Clean up
2376 | tg.cancel_scope.cancel()
2377 | await write_stream.aclose()
2378 | await read_stream.aclose()
2379 | ```
2380 | </Tab>
2381 | </Tabs>
2382 |
2383 | ## Error Handling
2384 |
2385 | Transport implementations should handle various error scenarios:
2386 |
2387 | 1. Connection errors
2388 | 2. Message parsing errors
2389 | 3. Protocol errors
2390 | 4. Network timeouts
2391 | 5. Resource cleanup
2392 |
2393 | Example error handling:
2394 |
2395 | <Tabs>
2396 | <Tab title="TypeScript">
2397 | ```typescript
2398 | class ExampleTransport implements Transport {
2399 | async start() {
2400 | try {
2401 | // Connection logic
2402 | } catch (error) {
2403 | this.onerror?.(new Error(`Failed to connect: ${error}`));
2404 | throw error;
2405 | }
2406 | }
2407 |
2408 | async send(message: JSONRPCMessage) {
2409 | try {
2410 | // Sending logic
2411 | } catch (error) {
2412 | this.onerror?.(new Error(`Failed to send message: ${error}`));
2413 | throw error;
2414 | }
2415 | }
2416 | }
2417 | ```
2418 | </Tab>
2419 |
2420 | <Tab title="Python">
2421 | Note that while MCP Servers are often implemented with asyncio, we recommend
2422 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2423 |
2424 | ```python
2425 | @contextmanager
2426 | async def example_transport(scope: Scope, receive: Receive, send: Send):
2427 | try:
2428 | # Create streams for bidirectional communication
2429 | read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
2430 | write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
2431 |
2432 | async def message_handler():
2433 | try:
2434 | async with read_stream_writer:
2435 | # Message handling logic
2436 | pass
2437 | except Exception as exc:
2438 | logger.error(f"Failed to handle message: {exc}")
2439 | raise exc
2440 |
2441 | async with anyio.create_task_group() as tg:
2442 | tg.start_soon(message_handler)
2443 | try:
2444 | # Yield streams for communication
2445 | yield read_stream, write_stream
2446 | except Exception as exc:
2447 | logger.error(f"Transport error: {exc}")
2448 | raise exc
2449 | finally:
2450 | tg.cancel_scope.cancel()
2451 | await write_stream.aclose()
2452 | await read_stream.aclose()
2453 | except Exception as exc:
2454 | logger.error(f"Failed to initialize transport: {exc}")
2455 | raise exc
2456 | ```
2457 | </Tab>
2458 | </Tabs>
2459 |
2460 | ## Best Practices
2461 |
2462 | When implementing or using MCP transport:
2463 |
2464 | 1. Handle connection lifecycle properly
2465 | 2. Implement proper error handling
2466 | 3. Clean up resources on connection close
2467 | 4. Use appropriate timeouts
2468 | 5. Validate messages before sending
2469 | 6. Log transport events for debugging
2470 | 7. Implement reconnection logic when appropriate
2471 | 8. Handle backpressure in message queues
2472 | 9. Monitor connection health
2473 | 10. Implement proper security measures
2474 |
2475 | ## Security Considerations
2476 |
2477 | When implementing transport:
2478 |
2479 | ### Authentication and Authorization
2480 |
2481 | * Implement proper authentication mechanisms
2482 | * Validate client credentials
2483 | * Use secure token handling
2484 | * Implement authorization checks
2485 |
2486 | ### Data Security
2487 |
2488 | * Use TLS for network transport
2489 | * Encrypt sensitive data
2490 | * Validate message integrity
2491 | * Implement message size limits
2492 | * Sanitize input data
2493 |
2494 | ### Network Security
2495 |
2496 | * Implement rate limiting
2497 | * Use appropriate timeouts
2498 | * Handle denial of service scenarios
2499 | * Monitor for unusual patterns
2500 | * Implement proper firewall rules
2501 |
2502 | ## Debugging Transport
2503 |
2504 | Tips for debugging transport issues:
2505 |
2506 | 1. Enable debug logging
2507 | 2. Monitor message flow
2508 | 3. Check connection states
2509 | 4. Validate message formats
2510 | 5. Test error scenarios
2511 | 6. Use network analysis tools
2512 | 7. Implement health checks
2513 | 8. Monitor resource usage
2514 | 9. Test edge cases
2515 | 10. Use proper error tracking
2516 |
2517 |
2518 | # Debugging
2519 | Source: https://modelcontextprotocol.io/docs/tools/debugging
2520 |
2521 | A comprehensive guide to debugging Model Context Protocol (MCP) integrations
2522 |
2523 | 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.
2524 |
2525 | <Info>
2526 | This guide is for macOS. Guides for other platforms are coming soon.
2527 | </Info>
2528 |
2529 | ## Debugging tools overview
2530 |
2531 | MCP provides several tools for debugging at different levels:
2532 |
2533 | 1. **MCP Inspector**
2534 | * Interactive debugging interface
2535 | * Direct server testing
2536 | * See the [Inspector guide](/docs/tools/inspector) for details
2537 |
2538 | 2. **Claude Desktop Developer Tools**
2539 | * Integration testing
2540 | * Log collection
2541 | * Chrome DevTools integration
2542 |
2543 | 3. **Server Logging**
2544 | * Custom logging implementations
2545 | * Error tracking
2546 | * Performance monitoring
2547 |
2548 | ## Debugging in Claude Desktop
2549 |
2550 | ### Checking server status
2551 |
2552 | The Claude.app interface provides basic server status information:
2553 |
2554 | 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:
2555 | * Connected servers
2556 | * Available prompts and resources
2557 |
2558 | 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:
2559 | * Tools made available to the model
2560 |
2561 | ### Viewing logs
2562 |
2563 | Review detailed MCP logs from Claude Desktop:
2564 |
2565 | ```bash
2566 | # Follow logs in real-time
2567 | tail -n 20 -F ~/Library/Logs/Claude/mcp*.log
2568 | ```
2569 |
2570 | The logs capture:
2571 |
2572 | * Server connection events
2573 | * Configuration issues
2574 | * Runtime errors
2575 | * Message exchanges
2576 |
2577 | ### Using Chrome DevTools
2578 |
2579 | Access Chrome's developer tools inside Claude Desktop to investigate client-side errors:
2580 |
2581 | 1. Create a `developer_settings.json` file with `allowDevTools` set to true:
2582 |
2583 | ```bash
2584 | echo '{"allowDevTools": true}' > ~/Library/Application\ Support/Claude/developer_settings.json
2585 | ```
2586 |
2587 | 2. Open DevTools: `Command-Option-Shift-i`
2588 |
2589 | Note: You'll see two DevTools windows:
2590 |
2591 | * Main content window
2592 | * App title bar window
2593 |
2594 | Use the Console panel to inspect client-side errors.
2595 |
2596 | Use the Network panel to inspect:
2597 |
2598 | * Message payloads
2599 | * Connection timing
2600 |
2601 | ## Common issues
2602 |
2603 | ### Working directory
2604 |
2605 | When using MCP servers with Claude Desktop:
2606 |
2607 | * 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
2608 | * Always use absolute paths in your configuration and `.env` files to ensure reliable operation
2609 | * For testing servers directly via command line, the working directory will be where you run the command
2610 |
2611 | For example in `claude_desktop_config.json`, use:
2612 |
2613 | ```json
2614 | {
2615 | "command": "npx",
2616 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/username/data"]
2617 | }
2618 | ```
2619 |
2620 | Instead of relative paths like `./data`
2621 |
2622 | ### Environment variables
2623 |
2624 | MCP servers inherit only a subset of environment variables automatically, like `USER`, `HOME`, and `PATH`.
2625 |
2626 | To override the default variables or provide your own, you can specify an `env` key in `claude_desktop_config.json`:
2627 |
2628 | ```json
2629 | {
2630 | "myserver": {
2631 | "command": "mcp-server-myapp",
2632 | "env": {
2633 | "MYAPP_API_KEY": "some_key",
2634 | }
2635 | }
2636 | }
2637 | ```
2638 |
2639 | ### Server initialization
2640 |
2641 | Common initialization problems:
2642 |
2643 | 1. **Path Issues**
2644 | * Incorrect server executable path
2645 | * Missing required files
2646 | * Permission problems
2647 | * Try using an absolute path for `command`
2648 |
2649 | 2. **Configuration Errors**
2650 | * Invalid JSON syntax
2651 | * Missing required fields
2652 | * Type mismatches
2653 |
2654 | 3. **Environment Problems**
2655 | * Missing environment variables
2656 | * Incorrect variable values
2657 | * Permission restrictions
2658 |
2659 | ### Connection problems
2660 |
2661 | When servers fail to connect:
2662 |
2663 | 1. Check Claude Desktop logs
2664 | 2. Verify server process is running
2665 | 3. Test standalone with [Inspector](/docs/tools/inspector)
2666 | 4. Verify protocol compatibility
2667 |
2668 | ## Implementing logging
2669 |
2670 | ### Server-side logging
2671 |
2672 | 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.
2673 |
2674 | <Warning>
2675 | Local MCP servers should not log messages to stdout (standard out), as this will interfere with protocol operation.
2676 | </Warning>
2677 |
2678 | For all [transports](/docs/concepts/transports), you can also provide logging to the client by sending a log message notification:
2679 |
2680 | <Tabs>
2681 | <Tab title="Python">
2682 | ```python
2683 | server.request_context.session.send_log_message(
2684 | level="info",
2685 | data="Server started successfully",
2686 | )
2687 | ```
2688 | </Tab>
2689 |
2690 | <Tab title="TypeScript">
2691 | ```typescript
2692 | server.sendLoggingMessage({
2693 | level: "info",
2694 | data: "Server started successfully",
2695 | });
2696 | ```
2697 | </Tab>
2698 | </Tabs>
2699 |
2700 | Important events to log:
2701 |
2702 | * Initialization steps
2703 | * Resource access
2704 | * Tool execution
2705 | * Error conditions
2706 | * Performance metrics
2707 |
2708 | ### Client-side logging
2709 |
2710 | In client applications:
2711 |
2712 | 1. Enable debug logging
2713 | 2. Monitor network traffic
2714 | 3. Track message exchanges
2715 | 4. Record error states
2716 |
2717 | ## Debugging workflow
2718 |
2719 | ### Development cycle
2720 |
2721 | 1. Initial Development
2722 | * Use [Inspector](/docs/tools/inspector) for basic testing
2723 | * Implement core functionality
2724 | * Add logging points
2725 |
2726 | 2. Integration Testing
2727 | * Test in Claude Desktop
2728 | * Monitor logs
2729 | * Check error handling
2730 |
2731 | ### Testing changes
2732 |
2733 | To test changes efficiently:
2734 |
2735 | * **Configuration changes**: Restart Claude Desktop
2736 | * **Server code changes**: Use Command-R to reload
2737 | * **Quick iteration**: Use [Inspector](/docs/tools/inspector) during development
2738 |
2739 | ## Best practices
2740 |
2741 | ### Logging strategy
2742 |
2743 | 1. **Structured Logging**
2744 | * Use consistent formats
2745 | * Include context
2746 | * Add timestamps
2747 | * Track request IDs
2748 |
2749 | 2. **Error Handling**
2750 | * Log stack traces
2751 | * Include error context
2752 | * Track error patterns
2753 | * Monitor recovery
2754 |
2755 | 3. **Performance Tracking**
2756 | * Log operation timing
2757 | * Monitor resource usage
2758 | * Track message sizes
2759 | * Measure latency
2760 |
2761 | ### Security considerations
2762 |
2763 | When debugging:
2764 |
2765 | 1. **Sensitive Data**
2766 | * Sanitize logs
2767 | * Protect credentials
2768 | * Mask personal information
2769 |
2770 | 2. **Access Control**
2771 | * Verify permissions
2772 | * Check authentication
2773 | * Monitor access patterns
2774 |
2775 | ## Getting help
2776 |
2777 | When encountering issues:
2778 |
2779 | 1. **First Steps**
2780 | * Check server logs
2781 | * Test with [Inspector](/docs/tools/inspector)
2782 | * Review configuration
2783 | * Verify environment
2784 |
2785 | 2. **Support Channels**
2786 | * GitHub issues
2787 | * GitHub discussions
2788 |
2789 | 3. **Providing Information**
2790 | * Log excerpts
2791 | * Configuration files
2792 | * Steps to reproduce
2793 | * Environment details
2794 |
2795 | ## Next steps
2796 |
2797 | <CardGroup cols={2}>
2798 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
2799 | Learn to use the MCP Inspector
2800 | </Card>
2801 | </CardGroup>
2802 |
2803 |
2804 | # Inspector
2805 | Source: https://modelcontextprotocol.io/docs/tools/inspector
2806 |
2807 | In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers
2808 |
2809 | 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.
2810 |
2811 | ## Getting started
2812 |
2813 | ### Installation and basic usage
2814 |
2815 | The Inspector runs directly through `npx` without requiring installation:
2816 |
2817 | ```bash
2818 | npx @modelcontextprotocol/inspector <command>
2819 | ```
2820 |
2821 | ```bash
2822 | npx @modelcontextprotocol/inspector <command> <arg1> <arg2>
2823 | ```
2824 |
2825 | #### Inspecting servers from NPM or PyPi
2826 |
2827 | A common way to start server packages from [NPM](https://npmjs.com) or [PyPi](https://pypi.com).
2828 |
2829 | <Tabs>
2830 | <Tab title="NPM package">
2831 | ```bash
2832 | npx -y @modelcontextprotocol/inspector npx <package-name> <args>
2833 | # For example
2834 | npx -y @modelcontextprotocol/inspector npx server-postgres postgres://127.0.0.1/testdb
2835 | ```
2836 | </Tab>
2837 |
2838 | <Tab title="PyPi package">
2839 | ```bash
2840 | npx @modelcontextprotocol/inspector uvx <package-name> <args>
2841 | # For example
2842 | npx @modelcontextprotocol/inspector uvx mcp-server-git --repository ~/code/mcp/servers.git
2843 | ```
2844 | </Tab>
2845 | </Tabs>
2846 |
2847 | #### Inspecting locally developed servers
2848 |
2849 | To inspect servers locally developed or downloaded as a repository, the most common
2850 | way is:
2851 |
2852 | <Tabs>
2853 | <Tab title="TypeScript">
2854 | ```bash
2855 | npx @modelcontextprotocol/inspector node path/to/server/index.js args...
2856 | ```
2857 | </Tab>
2858 |
2859 | <Tab title="Python">
2860 | ```bash
2861 | npx @modelcontextprotocol/inspector \
2862 | uv \
2863 | --directory path/to/server \
2864 | run \
2865 | package-name \
2866 | args...
2867 | ```
2868 | </Tab>
2869 | </Tabs>
2870 |
2871 | Please carefully read any attached README for the most accurate instructions.
2872 |
2873 | ## Feature overview
2874 |
2875 | <Frame caption="The MCP Inspector interface">
2876 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/mcp-inspector.png" />
2877 | </Frame>
2878 |
2879 | The Inspector provides several features for interacting with your MCP server:
2880 |
2881 | ### Server connection pane
2882 |
2883 | * Allows selecting the [transport](/docs/concepts/transports) for connecting to the server
2884 | * For local servers, supports customizing the command-line arguments and environment
2885 |
2886 | ### Resources tab
2887 |
2888 | * Lists all available resources
2889 | * Shows resource metadata (MIME types, descriptions)
2890 | * Allows resource content inspection
2891 | * Supports subscription testing
2892 |
2893 | ### Prompts tab
2894 |
2895 | * Displays available prompt templates
2896 | * Shows prompt arguments and descriptions
2897 | * Enables prompt testing with custom arguments
2898 | * Previews generated messages
2899 |
2900 | ### Tools tab
2901 |
2902 | * Lists available tools
2903 | * Shows tool schemas and descriptions
2904 | * Enables tool testing with custom inputs
2905 | * Displays tool execution results
2906 |
2907 | ### Notifications pane
2908 |
2909 | * Presents all logs recorded from the server
2910 | * Shows notifications received from the server
2911 |
2912 | ## Best practices
2913 |
2914 | ### Development workflow
2915 |
2916 | 1. Start Development
2917 | * Launch Inspector with your server
2918 | * Verify basic connectivity
2919 | * Check capability negotiation
2920 |
2921 | 2. Iterative testing
2922 | * Make server changes
2923 | * Rebuild the server
2924 | * Reconnect the Inspector
2925 | * Test affected features
2926 | * Monitor messages
2927 |
2928 | 3. Test edge cases
2929 | * Invalid inputs
2930 | * Missing prompt arguments
2931 | * Concurrent operations
2932 | * Verify error handling and error responses
2933 |
2934 | ## Next steps
2935 |
2936 | <CardGroup cols={2}>
2937 | <Card title="Inspector Repository" icon="github" href="https://github.com/modelcontextprotocol/inspector">
2938 | Check out the MCP Inspector source code
2939 | </Card>
2940 |
2941 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
2942 | Learn about broader debugging strategies
2943 | </Card>
2944 | </CardGroup>
2945 |
2946 |
2947 | # Example Servers
2948 | Source: https://modelcontextprotocol.io/examples
2949 |
2950 | A list of example servers and implementations
2951 |
2952 | 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.
2953 |
2954 | ## Reference implementations
2955 |
2956 | These official reference servers demonstrate core MCP features and SDK usage:
2957 |
2958 | ### Data and file systems
2959 |
2960 | * **[Filesystem](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)** - Secure file operations with configurable access controls
2961 | * **[PostgreSQL](https://github.com/modelcontextprotocol/servers/tree/main/src/postgres)** - Read-only database access with schema inspection capabilities
2962 | * **[SQLite](https://github.com/modelcontextprotocol/servers/tree/main/src/sqlite)** - Database interaction and business intelligence features
2963 | * **[Google Drive](https://github.com/modelcontextprotocol/servers/tree/main/src/gdrive)** - File access and search capabilities for Google Drive
2964 |
2965 | ### Development tools
2966 |
2967 | * **[Git](https://github.com/modelcontextprotocol/servers/tree/main/src/git)** - Tools to read, search, and manipulate Git repositories
2968 | * **[GitHub](https://github.com/modelcontextprotocol/servers/tree/main/src/github)** - Repository management, file operations, and GitHub API integration
2969 | * **[GitLab](https://github.com/modelcontextprotocol/servers/tree/main/src/gitlab)** - GitLab API integration enabling project management
2970 | * **[Sentry](https://github.com/modelcontextprotocol/servers/tree/main/src/sentry)** - Retrieving and analyzing issues from Sentry.io
2971 |
2972 | ### Web and browser automation
2973 |
2974 | * **[Brave Search](https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search)** - Web and local search using Brave's Search API
2975 | * **[Fetch](https://github.com/modelcontextprotocol/servers/tree/main/src/fetch)** - Web content fetching and conversion optimized for LLM usage
2976 | * **[Puppeteer](https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer)** - Browser automation and web scraping capabilities
2977 |
2978 | ### Productivity and communication
2979 |
2980 | * **[Slack](https://github.com/modelcontextprotocol/servers/tree/main/src/slack)** - Channel management and messaging capabilities
2981 | * **[Google Maps](https://github.com/modelcontextprotocol/servers/tree/main/src/google-maps)** - Location services, directions, and place details
2982 | * **[Memory](https://github.com/modelcontextprotocol/servers/tree/main/src/memory)** - Knowledge graph-based persistent memory system
2983 |
2984 | ### AI and specialized tools
2985 |
2986 | * **[EverArt](https://github.com/modelcontextprotocol/servers/tree/main/src/everart)** - AI image generation using various models
2987 | * **[Sequential Thinking](https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking)** - Dynamic problem-solving through thought sequences
2988 | * **[AWS KB Retrieval](https://github.com/modelcontextprotocol/servers/tree/main/src/aws-kb-retrieval-server)** - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime
2989 |
2990 | ## Official integrations
2991 |
2992 | These MCP servers are maintained by companies for their platforms:
2993 |
2994 | * **[Axiom](https://github.com/axiomhq/mcp-server-axiom)** - Query and analyze logs, traces, and event data using natural language
2995 | * **[Browserbase](https://github.com/browserbase/mcp-server-browserbase)** - Automate browser interactions in the cloud
2996 | * **[Cloudflare](https://github.com/cloudflare/mcp-server-cloudflare)** - Deploy and manage resources on the Cloudflare developer platform
2997 | * **[E2B](https://github.com/e2b-dev/mcp-server)** - Execute code in secure cloud sandboxes
2998 | * **[Neon](https://github.com/neondatabase/mcp-server-neon)** - Interact with the Neon serverless Postgres platform
2999 | * **[Obsidian Markdown Notes](https://github.com/calclavia/mcp-obsidian)** - Read and search through Markdown notes in Obsidian vaults
3000 | * **[Qdrant](https://github.com/qdrant/mcp-server-qdrant/)** - Implement semantic memory using the Qdrant vector search engine
3001 | * **[Raygun](https://github.com/MindscapeHQ/mcp-server-raygun)** - Access crash reporting and monitoring data
3002 | * **[Search1API](https://github.com/fatwang2/search1api-mcp)** - Unified API for search, crawling, and sitemaps
3003 | * **[Stripe](https://github.com/stripe/agent-toolkit)** - Interact with the Stripe API
3004 | * **[Tinybird](https://github.com/tinybirdco/mcp-tinybird)** - Interface with the Tinybird serverless ClickHouse platform
3005 |
3006 | ## Community highlights
3007 |
3008 | A growing ecosystem of community-developed servers extends MCP's capabilities:
3009 |
3010 | * **[Docker](https://github.com/ckreiling/mcp-server-docker)** - Manage containers, images, volumes, and networks
3011 | * **[Kubernetes](https://github.com/Flux159/mcp-server-kubernetes)** - Manage pods, deployments, and services
3012 | * **[Linear](https://github.com/jerhadf/linear-mcp-server)** - Project management and issue tracking
3013 | * **[Snowflake](https://github.com/datawiz168/mcp-snowflake-service)** - Interact with Snowflake databases
3014 | * **[Spotify](https://github.com/varunneal/spotify-mcp)** - Control Spotify playback and manage playlists
3015 | * **[Todoist](https://github.com/abhiz123/todoist-mcp-server)** - Task management integration
3016 |
3017 | > **Note:** Community servers are untested and should be used at your own risk. They are not affiliated with or endorsed by Anthropic.
3018 |
3019 | For a complete list of community servers, visit the [MCP Servers Repository](https://github.com/modelcontextprotocol/servers).
3020 |
3021 | ## Getting started
3022 |
3023 | ### Using reference servers
3024 |
3025 | TypeScript-based servers can be used directly with `npx`:
3026 |
3027 | ```bash
3028 | npx -y @modelcontextprotocol/server-memory
3029 | ```
3030 |
3031 | Python-based servers can be used with `uvx` (recommended) or `pip`:
3032 |
3033 | ```bash
3034 | # Using uvx
3035 | uvx mcp-server-git
3036 |
3037 | # Using pip
3038 | pip install mcp-server-git
3039 | python -m mcp_server_git
3040 | ```
3041 |
3042 | ### Configuring with Claude
3043 |
3044 | To use an MCP server with Claude, add it to your configuration:
3045 |
3046 | ```json
3047 | {
3048 | "mcpServers": {
3049 | "memory": {
3050 | "command": "npx",
3051 | "args": ["-y", "@modelcontextprotocol/server-memory"]
3052 | },
3053 | "filesystem": {
3054 | "command": "npx",
3055 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
3056 | },
3057 | "github": {
3058 | "command": "npx",
3059 | "args": ["-y", "@modelcontextprotocol/server-github"],
3060 | "env": {
3061 | "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
3062 | }
3063 | }
3064 | }
3065 | }
3066 | ```
3067 |
3068 | ## Additional resources
3069 |
3070 | * [MCP Servers Repository](https://github.com/modelcontextprotocol/servers) - Complete collection of reference implementations and community servers
3071 | * [Awesome MCP Servers](https://github.com/punkpeye/awesome-mcp-servers) - Curated list of MCP servers
3072 | * [MCP CLI](https://github.com/wong2/mcp-cli) - Command-line inspector for testing MCP servers
3073 | * [MCP Get](https://mcp-get.com) - Tool for installing and managing MCP servers
3074 | * [Supergateway](https://github.com/supercorp-ai/supergateway) - Run MCP stdio servers over SSE
3075 |
3076 | Visit our [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions) to engage with the MCP community.
3077 |
3078 |
3079 | # Introduction
3080 | Source: https://modelcontextprotocol.io/introduction
3081 |
3082 | Get started with the Model Context Protocol (MCP)
3083 |
3084 | <Note>Java SDK released! Check out [what else is new.](/development/updates)</Note>
3085 |
3086 | 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.
3087 |
3088 | ## Why MCP?
3089 |
3090 | MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:
3091 |
3092 | * A growing list of pre-built integrations that your LLM can directly plug into
3093 | * The flexibility to switch between LLM providers and vendors
3094 | * Best practices for securing your data within your infrastructure
3095 |
3096 | ### General architecture
3097 |
3098 | At its core, MCP follows a client-server architecture where a host application can connect to multiple servers:
3099 |
3100 | ```mermaid
3101 | flowchart LR
3102 | subgraph "Your Computer"
3103 | Host["Host with MCP Client\n(Claude, IDEs, Tools)"]
3104 | S1["MCP Server A"]
3105 | S2["MCP Server B"]
3106 | S3["MCP Server C"]
3107 | Host <-->|"MCP Protocol"| S1
3108 | Host <-->|"MCP Protocol"| S2
3109 | Host <-->|"MCP Protocol"| S3
3110 | S1 <--> D1[("Local\nData Source A")]
3111 | S2 <--> D2[("Local\nData Source B")]
3112 | end
3113 | subgraph "Internet"
3114 | S3 <-->|"Web APIs"| D3[("Remote\nService C")]
3115 | end
3116 | ```
3117 |
3118 | * **MCP Hosts**: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
3119 | * **MCP Clients**: Protocol clients that maintain 1:1 connections with servers
3120 | * **MCP Servers**: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
3121 | * **Local Data Sources**: Your computer's files, databases, and services that MCP servers can securely access
3122 | * **Remote Services**: External systems available over the internet (e.g., through APIs) that MCP servers can connect to
3123 |
3124 | ## Get started
3125 |
3126 | Choose the path that best fits your needs:
3127 |
3128 | #### Quick Starts
3129 |
3130 | <CardGroup cols={2}>
3131 | <Card title="For Server Developers" icon="bolt" href="/quickstart/server">
3132 | Get started building your own server to use in Claude for Desktop and other clients
3133 | </Card>
3134 |
3135 | <Card title="For Client Developers" icon="bolt" href="/quickstart/client">
3136 | Get started building your own client that can integrate with all MCP servers
3137 | </Card>
3138 |
3139 | <Card title="For Claude Desktop Users" icon="bolt" href="/quickstart/user">
3140 | Get started using pre-built servers in Claude for Desktop
3141 | </Card>
3142 | </CardGroup>
3143 |
3144 | #### Examples
3145 |
3146 | <CardGroup cols={2}>
3147 | <Card title="Example Servers" icon="grid" href="/examples">
3148 | Check out our gallery of official MCP servers and implementations
3149 | </Card>
3150 |
3151 | <Card title="Example Clients" icon="cubes" href="/clients">
3152 | View the list of clients that support MCP integrations
3153 | </Card>
3154 | </CardGroup>
3155 |
3156 | ## Tutorials
3157 |
3158 | <CardGroup cols={2}>
3159 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
3160 | Learn how to use LLMs like Claude to speed up your MCP development
3161 | </Card>
3162 |
3163 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
3164 | Learn how to effectively debug MCP servers and integrations
3165 | </Card>
3166 |
3167 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
3168 | Test and inspect your MCP servers with our interactive debugging tool
3169 | </Card>
3170 | </CardGroup>
3171 |
3172 | ## Explore MCP
3173 |
3174 | Dive deeper into MCP's core concepts and capabilities:
3175 |
3176 | <CardGroup cols={2}>
3177 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3178 | Understand how MCP connects clients, servers, and LLMs
3179 | </Card>
3180 |
3181 | <Card title="Resources" icon="database" href="/docs/concepts/resources">
3182 | Expose data and content from your servers to LLMs
3183 | </Card>
3184 |
3185 | <Card title="Prompts" icon="message" href="/docs/concepts/prompts">
3186 | Create reusable prompt templates and workflows
3187 | </Card>
3188 |
3189 | <Card title="Tools" icon="wrench" href="/docs/concepts/tools">
3190 | Enable LLMs to perform actions through your server
3191 | </Card>
3192 |
3193 | <Card title="Sampling" icon="robot" href="/docs/concepts/sampling">
3194 | Let your servers request completions from LLMs
3195 | </Card>
3196 |
3197 | <Card title="Transports" icon="network-wired" href="/docs/concepts/transports">
3198 | Learn about MCP's communication mechanism
3199 | </Card>
3200 | </CardGroup>
3201 |
3202 | ## Contributing
3203 |
3204 | Want to contribute? Check out our [Contributing Guide](/development/contributing) to learn how you can help improve MCP.
3205 |
3206 | ## Support and Feedback
3207 |
3208 | Here's how to get help or provide feedback:
3209 |
3210 | * 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)
3211 | * For discussions or Q\&A about the MCP specification, use the [specification discussions](https://github.com/modelcontextprotocol/specification/discussions)
3212 | * For discussions or Q\&A about other MCP open source components, use the [organization discussions](https://github.com/orgs/modelcontextprotocol/discussions)
3213 | * For bug reports, feature requests, and questions related to Claude.app and claude.ai's MCP integration, please email [[email protected]](mailto:[email protected])
3214 |
3215 |
3216 | # For Client Developers
3217 | Source: https://modelcontextprotocol.io/quickstart/client
3218 |
3219 | Get started building your own client that can integrate with all MCP servers.
3220 |
3221 | 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.
3222 |
3223 | <Tabs>
3224 | <Tab title="Python">
3225 | [You can find the complete code for this tutorial here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/mcp-client)
3226 |
3227 | ## System Requirements
3228 |
3229 | Before starting, ensure your system meets these requirements:
3230 |
3231 | * Mac or Windows computer
3232 | * Latest Python version installed
3233 | * Latest version of `uv` installed
3234 |
3235 | ## Setting Up Your Environment
3236 |
3237 | First, create a new Python project with `uv`:
3238 |
3239 | ```bash
3240 | # Create project directory
3241 | uv init mcp-client
3242 | cd mcp-client
3243 |
3244 | # Create virtual environment
3245 | uv venv
3246 |
3247 | # Activate virtual environment
3248 | # On Windows:
3249 | .venv\Scripts\activate
3250 | # On Unix or MacOS:
3251 | source .venv/bin/activate
3252 |
3253 | # Install required packages
3254 | uv add mcp anthropic python-dotenv
3255 |
3256 | # Remove boilerplate files
3257 | rm hello.py
3258 |
3259 | # Create our main file
3260 | touch client.py
3261 | ```
3262 |
3263 | ## Setting Up Your API Key
3264 |
3265 | You'll need an Anthropic API key from the [Anthropic Console](https://console.anthropic.com/settings/keys).
3266 |
3267 | Create a `.env` file to store it:
3268 |
3269 | ```bash
3270 | # Create .env file
3271 | touch .env
3272 | ```
3273 |
3274 | Add your key to the `.env` file:
3275 |
3276 | ```bash
3277 | ANTHROPIC_API_KEY=<your key here>
3278 | ```
3279 |
3280 | Add `.env` to your `.gitignore`:
3281 |
3282 | ```bash
3283 | echo ".env" >> .gitignore
3284 | ```
3285 |
3286 | <Warning>
3287 | Make sure you keep your `ANTHROPIC_API_KEY` secure!
3288 | </Warning>
3289 |
3290 | ## Creating the Client
3291 |
3292 | ### Basic Client Structure
3293 |
3294 | First, let's set up our imports and create the basic client class:
3295 |
3296 | ```python
3297 | import asyncio
3298 | from typing import Optional
3299 | from contextlib import AsyncExitStack
3300 |
3301 | from mcp import ClientSession, StdioServerParameters
3302 | from mcp.client.stdio import stdio_client
3303 |
3304 | from anthropic import Anthropic
3305 | from dotenv import load_dotenv
3306 |
3307 | load_dotenv() # load environment variables from .env
3308 |
3309 | class MCPClient:
3310 | def __init__(self):
3311 | # Initialize session and client objects
3312 | self.session: Optional[ClientSession] = None
3313 | self.exit_stack = AsyncExitStack()
3314 | self.anthropic = Anthropic()
3315 | # methods will go here
3316 | ```
3317 |
3318 | ### Server Connection Management
3319 |
3320 | Next, we'll implement the method to connect to an MCP server:
3321 |
3322 | ```python
3323 | async def connect_to_server(self, server_script_path: str):
3324 | """Connect to an MCP server
3325 |
3326 | Args:
3327 | server_script_path: Path to the server script (.py or .js)
3328 | """
3329 | is_python = server_script_path.endswith('.py')
3330 | is_js = server_script_path.endswith('.js')
3331 | if not (is_python or is_js):
3332 | raise ValueError("Server script must be a .py or .js file")
3333 |
3334 | command = "python" if is_python else "node"
3335 | server_params = StdioServerParameters(
3336 | command=command,
3337 | args=[server_script_path],
3338 | env=None
3339 | )
3340 |
3341 | stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
3342 | self.stdio, self.write = stdio_transport
3343 | self.session = await self.exit_stack.enter_async_context(ClientSession(self.stdio, self.write))
3344 |
3345 | await self.session.initialize()
3346 |
3347 | # List available tools
3348 | response = await self.session.list_tools()
3349 | tools = response.tools
3350 | print("\nConnected to server with tools:", [tool.name for tool in tools])
3351 | ```
3352 |
3353 | ### Query Processing Logic
3354 |
3355 | Now let's add the core functionality for processing queries and handling tool calls:
3356 |
3357 | ```python
3358 | async def process_query(self, query: str) -> str:
3359 | """Process a query using Claude and available tools"""
3360 | messages = [
3361 | {
3362 | "role": "user",
3363 | "content": query
3364 | }
3365 | ]
3366 |
3367 | response = await self.session.list_tools()
3368 | available_tools = [{
3369 | "name": tool.name,
3370 | "description": tool.description,
3371 | "input_schema": tool.inputSchema
3372 | } for tool in response.tools]
3373 |
3374 | # Initial Claude API call
3375 | response = self.anthropic.messages.create(
3376 | model="claude-3-5-sonnet-20241022",
3377 | max_tokens=1000,
3378 | messages=messages,
3379 | tools=available_tools
3380 | )
3381 |
3382 | # Process response and handle tool calls
3383 | tool_results = []
3384 | final_text = []
3385 |
3386 | assistant_message_content = []
3387 | for content in response.content:
3388 | if content.type == 'text':
3389 | final_text.append(content.text)
3390 | assistant_message_content.append(content)
3391 | elif content.type == 'tool_use':
3392 | tool_name = content.name
3393 | tool_args = content.input
3394 |
3395 | # Execute tool call
3396 | result = await self.session.call_tool(tool_name, tool_args)
3397 | tool_results.append({"call": tool_name, "result": result})
3398 | final_text.append(f"[Calling tool {tool_name} with args {tool_args}]")
3399 |
3400 | assistant_message_content.append(content)
3401 | messages.append({
3402 | "role": "assistant",
3403 | "content": assistant_message_content
3404 | })
3405 | messages.append({
3406 | "role": "user",
3407 | "content": [
3408 | {
3409 | "type": "tool_result",
3410 | "tool_use_id": content.id,
3411 | "content": result.content
3412 | }
3413 | ]
3414 | })
3415 |
3416 | # Get next response from Claude
3417 | response = self.anthropic.messages.create(
3418 | model="claude-3-5-sonnet-20241022",
3419 | max_tokens=1000,
3420 | messages=messages,
3421 | tools=available_tools
3422 | )
3423 |
3424 | final_text.append(response.content[0].text)
3425 |
3426 | return "\n".join(final_text)
3427 | ```
3428 |
3429 | ### Interactive Chat Interface
3430 |
3431 | Now we'll add the chat loop and cleanup functionality:
3432 |
3433 | ```python
3434 | async def chat_loop(self):
3435 | """Run an interactive chat loop"""
3436 | print("\nMCP Client Started!")
3437 | print("Type your queries or 'quit' to exit.")
3438 |
3439 | while True:
3440 | try:
3441 | query = input("\nQuery: ").strip()
3442 |
3443 | if query.lower() == 'quit':
3444 | break
3445 |
3446 | response = await self.process_query(query)
3447 | print("\n" + response)
3448 |
3449 | except Exception as e:
3450 | print(f"\nError: {str(e)}")
3451 |
3452 | async def cleanup(self):
3453 | """Clean up resources"""
3454 | await self.exit_stack.aclose()
3455 | ```
3456 |
3457 | ### Main Entry Point
3458 |
3459 | Finally, we'll add the main execution logic:
3460 |
3461 | ```python
3462 | async def main():
3463 | if len(sys.argv) < 2:
3464 | print("Usage: python client.py <path_to_server_script>")
3465 | sys.exit(1)
3466 |
3467 | client = MCPClient()
3468 | try:
3469 | await client.connect_to_server(sys.argv[1])
3470 | await client.chat_loop()
3471 | finally:
3472 | await client.cleanup()
3473 |
3474 | if __name__ == "__main__":
3475 | import sys
3476 | asyncio.run(main())
3477 | ```
3478 |
3479 | You can find the complete `client.py` file [here.](https://gist.github.com/zckly/f3f28ea731e096e53b39b47bf0a2d4b1)
3480 |
3481 | ## Key Components Explained
3482 |
3483 | ### 1. Client Initialization
3484 |
3485 | * The `MCPClient` class initializes with session management and API clients
3486 | * Uses `AsyncExitStack` for proper resource management
3487 | * Configures the Anthropic client for Claude interactions
3488 |
3489 | ### 2. Server Connection
3490 |
3491 | * Supports both Python and Node.js servers
3492 | * Validates server script type
3493 | * Sets up proper communication channels
3494 | * Initializes the session and lists available tools
3495 |
3496 | ### 3. Query Processing
3497 |
3498 | * Maintains conversation context
3499 | * Handles Claude's responses and tool calls
3500 | * Manages the message flow between Claude and tools
3501 | * Combines results into a coherent response
3502 |
3503 | ### 4. Interactive Interface
3504 |
3505 | * Provides a simple command-line interface
3506 | * Handles user input and displays responses
3507 | * Includes basic error handling
3508 | * Allows graceful exit
3509 |
3510 | ### 5. Resource Management
3511 |
3512 | * Proper cleanup of resources
3513 | * Error handling for connection issues
3514 | * Graceful shutdown procedures
3515 |
3516 | ## Common Customization Points
3517 |
3518 | 1. **Tool Handling**
3519 | * Modify `process_query()` to handle specific tool types
3520 | * Add custom error handling for tool calls
3521 | * Implement tool-specific response formatting
3522 |
3523 | 2. **Response Processing**
3524 | * Customize how tool results are formatted
3525 | * Add response filtering or transformation
3526 | * Implement custom logging
3527 |
3528 | 3. **User Interface**
3529 | * Add a GUI or web interface
3530 | * Implement rich console output
3531 | * Add command history or auto-completion
3532 |
3533 | ## Running the Client
3534 |
3535 | To run your client with any MCP server:
3536 |
3537 | ```bash
3538 | uv run client.py path/to/server.py # python server
3539 | uv run client.py path/to/build/index.js # node server
3540 | ```
3541 |
3542 | <Note>
3543 | 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`
3544 | </Note>
3545 |
3546 | The client will:
3547 |
3548 | 1. Connect to the specified server
3549 | 2. List available tools
3550 | 3. Start an interactive chat session where you can:
3551 | * Enter queries
3552 | * See tool executions
3553 | * Get responses from Claude
3554 |
3555 | Here's an example of what it should look like if connected to the weather server from the server quickstart:
3556 |
3557 | <Frame>
3558 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/client-claude-cli-python.png" />
3559 | </Frame>
3560 |
3561 | ## How It Works
3562 |
3563 | When you submit a query:
3564 |
3565 | 1. The client gets the list of available tools from the server
3566 | 2. Your query is sent to Claude along with tool descriptions
3567 | 3. Claude decides which tools (if any) to use
3568 | 4. The client executes any requested tool calls through the server
3569 | 5. Results are sent back to Claude
3570 | 6. Claude provides a natural language response
3571 | 7. The response is displayed to you
3572 |
3573 | ## Best practices
3574 |
3575 | 1. **Error Handling**
3576 | * Always wrap tool calls in try-catch blocks
3577 | * Provide meaningful error messages
3578 | * Gracefully handle connection issues
3579 |
3580 | 2. **Resource Management**
3581 | * Use `AsyncExitStack` for proper cleanup
3582 | * Close connections when done
3583 | * Handle server disconnections
3584 |
3585 | 3. **Security**
3586 | * Store API keys securely in `.env`
3587 | * Validate server responses
3588 | * Be cautious with tool permissions
3589 |
3590 | ## Troubleshooting
3591 |
3592 | ### Server Path Issues
3593 |
3594 | * Double-check the path to your server script is correct
3595 | * Use the absolute path if the relative path isn't working
3596 | * For Windows users, make sure to use forward slashes (/) or escaped backslashes (\\) in the path
3597 | * Verify the server file has the correct extension (.py for Python or .js for Node.js)
3598 |
3599 | Example of correct path usage:
3600 |
3601 | ```bash
3602 | # Relative path
3603 | uv run client.py ./server/weather.py
3604 |
3605 | # Absolute path
3606 | uv run client.py /Users/username/projects/mcp-server/weather.py
3607 |
3608 | # Windows path (either format works)
3609 | uv run client.py C:/projects/mcp-server/weather.py
3610 | uv run client.py C:\\projects\\mcp-server\\weather.py
3611 | ```
3612 |
3613 | ### Response Timing
3614 |
3615 | * The first response might take up to 30 seconds to return
3616 | * This is normal and happens while:
3617 | * The server initializes
3618 | * Claude processes the query
3619 | * Tools are being executed
3620 | * Subsequent responses are typically faster
3621 | * Don't interrupt the process during this initial waiting period
3622 |
3623 | ### Common Error Messages
3624 |
3625 | If you see:
3626 |
3627 | * `FileNotFoundError`: Check your server path
3628 | * `Connection refused`: Ensure the server is running and the path is correct
3629 | * `Tool execution failed`: Verify the tool's required environment variables are set
3630 | * `Timeout error`: Consider increasing the timeout in your client configuration
3631 | </Tab>
3632 |
3633 | <Tab title="Java">
3634 | <Note>
3635 | This is a quickstart demo based on Spring AI MCP auto-configuration and boot starters.
3636 | To learn how to create sync and async MCP Clients manually, consult the [Java SDK Client](/sdk/java/mcp-client) documentation
3637 | </Note>
3638 |
3639 | 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.
3640 | [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)
3641 |
3642 | ## System Requirements
3643 |
3644 | Before starting, ensure your system meets these requirements:
3645 |
3646 | * Java 17 or higher
3647 | * Maven 3.6+
3648 | * npx package manager
3649 | * Anthropic API key (Claude)
3650 | * Brave Search API key
3651 |
3652 | ## Setting Up Your Environment
3653 |
3654 | 1. Install npx (Node Package eXecute):
3655 | First, make sure to install [npm](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm)
3656 | and then run:
3657 | ```bash
3658 | npm install -g npx
3659 | ```
3660 |
3661 | 2. Clone the repository:
3662 | ```bash
3663 | git clone https://github.com/spring-projects/spring-ai-examples.git
3664 | cd model-context-protocol/brave-chatbot
3665 | ```
3666 |
3667 | 3. Set up your API keys:
3668 | ```bash
3669 | export ANTHROPIC_API_KEY='your-anthropic-api-key-here'
3670 | export BRAVE_API_KEY='your-brave-api-key-here'
3671 | ```
3672 |
3673 | 4. Build the application:
3674 | ```bash
3675 | ./mvnw clean install
3676 | ```
3677 |
3678 | 5. Run the application using Maven:
3679 | ```bash
3680 | ./mvnw spring-boot:run
3681 | ```
3682 |
3683 | <Warning>
3684 | Make sure you keep your `ANTHROPIC_API_KEY` and `BRAVE_API_KEY` keys secure!
3685 | </Warning>
3686 |
3687 | ## How it Works
3688 |
3689 | The application integrates Spring AI with the Brave Search MCP server through several components:
3690 |
3691 | ### MCP Client Configuration
3692 |
3693 | 1. Required dependencies in pom.xml:
3694 |
3695 | ```xml
3696 | <dependency>
3697 | <groupId>org.springframework.ai</groupId>
3698 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
3699 | </dependency>
3700 | <dependency>
3701 | <groupId>org.springframework.ai</groupId>
3702 | <artifactId>spring-ai-anthropic-spring-boot-starter</artifactId>
3703 | </dependency>
3704 | ```
3705 |
3706 | 2. Application properties (application.yml):
3707 |
3708 | ```yml
3709 | spring:
3710 | ai:
3711 | mcp:
3712 | client:
3713 | enabled: true
3714 | name: brave-search-client
3715 | version: 1.0.0
3716 | type: SYNC
3717 | request-timeout: 20s
3718 | stdio:
3719 | root-change-notification: true
3720 | servers-configuration: classpath:/mcp-servers-config.json
3721 | anthropic:
3722 | api-key: ${ANTHROPIC_API_KEY}
3723 | ```
3724 |
3725 | This activates the `spring-ai-mcp-client-spring-boot-starter` to create one or more `McpClient`s based on the provided server configuration.
3726 |
3727 | 3. MCP Server Configuration (`mcp-servers-config.json`):
3728 |
3729 | ```json
3730 | {
3731 | "mcpServers": {
3732 | "brave-search": {
3733 | "command": "npx",
3734 | "args": [
3735 | "-y",
3736 | "@modelcontextprotocol/server-brave-search"
3737 | ],
3738 | "env": {
3739 | "BRAVE_API_KEY": "<PUT YOUR BRAVE API KEY>"
3740 | }
3741 | }
3742 | }
3743 | }
3744 | ```
3745 |
3746 | ### Chat Implementation
3747 |
3748 | The chatbot is implemented using Spring AI's ChatClient with MCP tool integration:
3749 |
3750 | ```java
3751 | var chatClient = chatClientBuilder
3752 | .defaultSystem("You are useful assistant, expert in AI and Java.")
3753 | .defaultTools((Object[]) mcpToolAdapter.toolCallbacks())
3754 | .defaultAdvisors(new MessageChatMemoryAdvisor(new InMemoryChatMemory()))
3755 | .build();
3756 | ```
3757 |
3758 | Key features:
3759 |
3760 | * Uses Claude AI model for natural language understanding
3761 | * Integrates Brave Search through MCP for real-time web search capabilities
3762 | * Maintains conversation memory using InMemoryChatMemory
3763 | * Runs as an interactive command-line application
3764 |
3765 | ### Build and run
3766 |
3767 | ```bash
3768 | ./mvnw clean install
3769 | java -jar ./target/ai-mcp-brave-chatbot-0.0.1-SNAPSHOT.jar
3770 | ```
3771 |
3772 | or
3773 |
3774 | ```bash
3775 | ./mvnw spring-boot:run
3776 | ```
3777 |
3778 | 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.
3779 |
3780 | The chatbot can:
3781 |
3782 | * Answer questions using its built-in knowledge
3783 | * Perform web searches when needed using Brave Search
3784 | * Remember context from previous messages in the conversation
3785 | * Combine information from multiple sources to provide comprehensive answers
3786 |
3787 | ### Advanced Configuration
3788 |
3789 | The MCP client supports additional configuration options:
3790 |
3791 | * Client customization through `McpSyncClientCustomizer` or `McpAsyncClientCustomizer`
3792 | * Multiple clients with multiple transport types: `STDIO` and `SSE` (Server-Sent Events)
3793 | * Integration with Spring AI's tool execution framework
3794 | * Automatic client initialization and lifecycle management
3795 |
3796 | For WebFlux-based applications, you can use the WebFlux starter instead:
3797 |
3798 | ```xml
3799 | <dependency>
3800 | <groupId>org.springframework.ai</groupId>
3801 | <artifactId>spring-ai-mcp-client-webflux-spring-boot-starter</artifactId>
3802 | </dependency>
3803 | ```
3804 |
3805 | This provides similar functionality but uses a WebFlux-based SSE transport implementation, recommended for production deployments.
3806 | </Tab>
3807 | </Tabs>
3808 |
3809 | ## Next steps
3810 |
3811 | <CardGroup cols={2}>
3812 | <Card title="Example servers" icon="grid" href="/examples">
3813 | Check out our gallery of official MCP servers and implementations
3814 | </Card>
3815 |
3816 | <Card title="Clients" icon="cubes" href="/clients">
3817 | View the list of clients that support MCP integrations
3818 | </Card>
3819 |
3820 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
3821 | Learn how to use LLMs like Claude to speed up your MCP development
3822 | </Card>
3823 |
3824 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3825 | Understand how MCP connects clients, servers, and LLMs
3826 | </Card>
3827 | </CardGroup>
3828 |
3829 |
3830 | # For Server Developers
3831 | Source: https://modelcontextprotocol.io/quickstart/server
3832 |
3833 | Get started building your own server to use in Claude for Desktop and other clients.
3834 |
3835 | 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.
3836 |
3837 | ### What we'll be building
3838 |
3839 | 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!
3840 |
3841 | 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):
3842 |
3843 | <Frame>
3844 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
3845 | </Frame>
3846 |
3847 | <Frame>
3848 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
3849 | </Frame>
3850 |
3851 | <Note>
3852 | 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).
3853 | </Note>
3854 |
3855 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
3856 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
3857 | </Accordion>
3858 |
3859 | ### Core MCP Concepts
3860 |
3861 | MCP servers can provide three main types of capabilities:
3862 |
3863 | 1. **Resources**: File-like data that can be read by clients (like API responses or file contents)
3864 | 2. **Tools**: Functions that can be called by the LLM (with user approval)
3865 | 3. **Prompts**: Pre-written templates that help users accomplish specific tasks
3866 |
3867 | This tutorial will primarily focus on tools.
3868 |
3869 | <Tabs>
3870 | <Tab title="Python">
3871 | 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)
3872 |
3873 | ### Prerequisite knowledge
3874 |
3875 | This quickstart assumes you have familiarity with:
3876 |
3877 | * Python
3878 | * LLMs like Claude
3879 |
3880 | ### System requirements
3881 |
3882 | * Python 3.10 or higher installed.
3883 | * You must use the Python MCP SDK 1.2.0 or higher.
3884 |
3885 | ### Set up your environment
3886 |
3887 | First, let's install `uv` and set up our Python project and environment:
3888 |
3889 | <CodeGroup>
3890 | ```bash MacOS/Linux
3891 | curl -LsSf https://astral.sh/uv/install.sh | sh
3892 | ```
3893 |
3894 | ```powershell Windows
3895 | powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
3896 | ```
3897 | </CodeGroup>
3898 |
3899 | Make sure to restart your terminal afterwards to ensure that the `uv` command gets picked up.
3900 |
3901 | Now, let's create and set up our project:
3902 |
3903 | <CodeGroup>
3904 | ```bash MacOS/Linux
3905 | # Create a new directory for our project
3906 | uv init weather
3907 | cd weather
3908 |
3909 | # Create virtual environment and activate it
3910 | uv venv
3911 | source .venv/bin/activate
3912 |
3913 | # Install dependencies
3914 | uv add "mcp[cli]" httpx
3915 |
3916 | # Create our server file
3917 | touch weather.py
3918 | ```
3919 |
3920 | ```powershell Windows
3921 | # Create a new directory for our project
3922 | uv init weather
3923 | cd weather
3924 |
3925 | # Create virtual environment and activate it
3926 | uv venv
3927 | .venv\Scripts\activate
3928 |
3929 | # Install dependencies
3930 | uv add mcp[cli] httpx
3931 |
3932 | # Create our server file
3933 | new-item weather.py
3934 | ```
3935 | </CodeGroup>
3936 |
3937 | Now let's dive into building your server.
3938 |
3939 | ## Building your server
3940 |
3941 | ### Importing packages and setting up the instance
3942 |
3943 | Add these to the top of your `weather.py`:
3944 |
3945 | ```python
3946 | from typing import Any
3947 | import httpx
3948 | from mcp.server.fastmcp import FastMCP
3949 |
3950 | # Initialize FastMCP server
3951 | mcp = FastMCP("weather")
3952 |
3953 | # Constants
3954 | NWS_API_BASE = "https://api.weather.gov"
3955 | USER_AGENT = "weather-app/1.0"
3956 | ```
3957 |
3958 | The FastMCP class uses Python type hints and docstrings to automatically generate tool definitions, making it easy to create and maintain MCP tools.
3959 |
3960 | ### Helper functions
3961 |
3962 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
3963 |
3964 | ```python
3965 | async def make_nws_request(url: str) -> dict[str, Any] | None:
3966 | """Make a request to the NWS API with proper error handling."""
3967 | headers = {
3968 | "User-Agent": USER_AGENT,
3969 | "Accept": "application/geo+json"
3970 | }
3971 | async with httpx.AsyncClient() as client:
3972 | try:
3973 | response = await client.get(url, headers=headers, timeout=30.0)
3974 | response.raise_for_status()
3975 | return response.json()
3976 | except Exception:
3977 | return None
3978 |
3979 | def format_alert(feature: dict) -> str:
3980 | """Format an alert feature into a readable string."""
3981 | props = feature["properties"]
3982 | return f"""
3983 | Event: {props.get('event', 'Unknown')}
3984 | Area: {props.get('areaDesc', 'Unknown')}
3985 | Severity: {props.get('severity', 'Unknown')}
3986 | Description: {props.get('description', 'No description available')}
3987 | Instructions: {props.get('instruction', 'No specific instructions provided')}
3988 | """
3989 | ```
3990 |
3991 | ### Implementing tool execution
3992 |
3993 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
3994 |
3995 | ```python
3996 | @mcp.tool()
3997 | async def get_alerts(state: str) -> str:
3998 | """Get weather alerts for a US state.
3999 |
4000 | Args:
4001 | state: Two-letter US state code (e.g. CA, NY)
4002 | """
4003 | url = f"{NWS_API_BASE}/alerts/active/area/{state}"
4004 | data = await make_nws_request(url)
4005 |
4006 | if not data or "features" not in data:
4007 | return "Unable to fetch alerts or no alerts found."
4008 |
4009 | if not data["features"]:
4010 | return "No active alerts for this state."
4011 |
4012 | alerts = [format_alert(feature) for feature in data["features"]]
4013 | return "\n---\n".join(alerts)
4014 |
4015 | @mcp.tool()
4016 | async def get_forecast(latitude: float, longitude: float) -> str:
4017 | """Get weather forecast for a location.
4018 |
4019 | Args:
4020 | latitude: Latitude of the location
4021 | longitude: Longitude of the location
4022 | """
4023 | # First get the forecast grid endpoint
4024 | points_url = f"{NWS_API_BASE}/points/{latitude},{longitude}"
4025 | points_data = await make_nws_request(points_url)
4026 |
4027 | if not points_data:
4028 | return "Unable to fetch forecast data for this location."
4029 |
4030 | # Get the forecast URL from the points response
4031 | forecast_url = points_data["properties"]["forecast"]
4032 | forecast_data = await make_nws_request(forecast_url)
4033 |
4034 | if not forecast_data:
4035 | return "Unable to fetch detailed forecast."
4036 |
4037 | # Format the periods into a readable forecast
4038 | periods = forecast_data["properties"]["periods"]
4039 | forecasts = []
4040 | for period in periods[:5]: # Only show next 5 periods
4041 | forecast = f"""
4042 | {period['name']}:
4043 | Temperature: {period['temperature']}°{period['temperatureUnit']}
4044 | Wind: {period['windSpeed']} {period['windDirection']}
4045 | Forecast: {period['detailedForecast']}
4046 | """
4047 | forecasts.append(forecast)
4048 |
4049 | return "\n---\n".join(forecasts)
4050 | ```
4051 |
4052 | ### Running the server
4053 |
4054 | Finally, let's initialize and run the server:
4055 |
4056 | ```python
4057 | if __name__ == "__main__":
4058 | # Initialize and run the server
4059 | mcp.run(transport='stdio')
4060 | ```
4061 |
4062 | Your server is complete! Run `uv run weather.py` to confirm that everything's working.
4063 |
4064 | Let's now test your server from an existing MCP host, Claude for Desktop.
4065 |
4066 | ## Testing your server with Claude for Desktop
4067 |
4068 | <Note>
4069 | 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.
4070 | </Note>
4071 |
4072 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4073 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4074 |
4075 | 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.
4076 |
4077 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4078 |
4079 | <Tabs>
4080 | <Tab title="MacOS/Linux">
4081 | ```bash
4082 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4083 | ```
4084 | </Tab>
4085 |
4086 | <Tab title="Windows">
4087 | ```powershell
4088 | code $env:AppData\Claude\claude_desktop_config.json
4089 | ```
4090 | </Tab>
4091 | </Tabs>
4092 |
4093 | 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.
4094 |
4095 | In this case, we'll add our single weather server like so:
4096 |
4097 | <Tabs>
4098 | <Tab title="MacOS/Linux">
4099 | ```json Python
4100 | {
4101 | "mcpServers": {
4102 | "weather": {
4103 | "command": "uv",
4104 | "args": [
4105 | "--directory",
4106 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather",
4107 | "run",
4108 | "weather.py"
4109 | ]
4110 | }
4111 | }
4112 | }
4113 | ```
4114 | </Tab>
4115 |
4116 | <Tab title="Windows">
4117 | ```json Python
4118 | {
4119 | "mcpServers": {
4120 | "weather": {
4121 | "command": "uv",
4122 | "args": [
4123 | "--directory",
4124 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather",
4125 | "run",
4126 | "weather.py"
4127 | ]
4128 | }
4129 | }
4130 | }
4131 | ```
4132 | </Tab>
4133 | </Tabs>
4134 |
4135 | <Warning>
4136 | 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.
4137 | </Warning>
4138 |
4139 | <Note>
4140 | Make sure you pass in the absolute path to your server.
4141 | </Note>
4142 |
4143 | This tells Claude for Desktop:
4144 |
4145 | 1. There's an MCP server named "weather"
4146 | 2. To launch it by running `uv --directory /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather run weather`
4147 |
4148 | Save the file, and restart **Claude for Desktop**.
4149 | </Tab>
4150 |
4151 | <Tab title="Node">
4152 | 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)
4153 |
4154 | ### Prerequisite knowledge
4155 |
4156 | This quickstart assumes you have familiarity with:
4157 |
4158 | * TypeScript
4159 | * LLMs like Claude
4160 |
4161 | ### System requirements
4162 |
4163 | For TypeScript, make sure you have the latest version of Node installed.
4164 |
4165 | ### Set up your environment
4166 |
4167 | First, let's install Node.js and npm if you haven't already. You can download them from [nodejs.org](https://nodejs.org/).
4168 | Verify your Node.js installation:
4169 |
4170 | ```bash
4171 | node --version
4172 | npm --version
4173 | ```
4174 |
4175 | For this tutorial, you'll need Node.js version 16 or higher.
4176 |
4177 | Now, let's create and set up our project:
4178 |
4179 | <CodeGroup>
4180 | ```bash MacOS/Linux
4181 | # Create a new directory for our project
4182 | mkdir weather
4183 | cd weather
4184 |
4185 | # Initialize a new npm project
4186 | npm init -y
4187 |
4188 | # Install dependencies
4189 | npm install @modelcontextprotocol/sdk zod
4190 | npm install -D @types/node typescript
4191 |
4192 | # Create our files
4193 | mkdir src
4194 | touch src/index.ts
4195 | ```
4196 |
4197 | ```powershell Windows
4198 | # Create a new directory for our project
4199 | md weather
4200 | cd weather
4201 |
4202 | # Initialize a new npm project
4203 | npm init -y
4204 |
4205 | # Install dependencies
4206 | npm install @modelcontextprotocol/sdk zod
4207 | npm install -D @types/node typescript
4208 |
4209 | # Create our files
4210 | md src
4211 | new-item src\index.ts
4212 | ```
4213 | </CodeGroup>
4214 |
4215 | Update your package.json to add type: "module" and a build script:
4216 |
4217 | ```json package.json
4218 | {
4219 | "type": "module",
4220 | "bin": {
4221 | "weather": "./build/index.js"
4222 | },
4223 | "scripts": {
4224 | "build": "tsc && node -e \"require('fs').chmodSync('build/index.js', '755')\"",
4225 | },
4226 | "files": [
4227 | "build"
4228 | ],
4229 | }
4230 | ```
4231 |
4232 | Create a `tsconfig.json` in the root of your project:
4233 |
4234 | ```json tsconfig.json
4235 | {
4236 | "compilerOptions": {
4237 | "target": "ES2022",
4238 | "module": "Node16",
4239 | "moduleResolution": "Node16",
4240 | "outDir": "./build",
4241 | "rootDir": "./src",
4242 | "strict": true,
4243 | "esModuleInterop": true,
4244 | "skipLibCheck": true,
4245 | "forceConsistentCasingInFileNames": true
4246 | },
4247 | "include": ["src/**/*"],
4248 | "exclude": ["node_modules"]
4249 | }
4250 | ```
4251 |
4252 | Now let's dive into building your server.
4253 |
4254 | ## Building your server
4255 |
4256 | ### Importing packages and setting up the instance
4257 |
4258 | Add these to the top of your `src/index.ts`:
4259 |
4260 | ```typescript
4261 | import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
4262 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
4263 | import { z } from "zod";
4264 |
4265 | const NWS_API_BASE = "https://api.weather.gov";
4266 | const USER_AGENT = "weather-app/1.0";
4267 |
4268 | // Create server instance
4269 | const server = new McpServer({
4270 | name: "weather",
4271 | version: "1.0.0",
4272 | });
4273 | ```
4274 |
4275 | ### Helper functions
4276 |
4277 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
4278 |
4279 | ```typescript
4280 | // Helper function for making NWS API requests
4281 | async function makeNWSRequest<T>(url: string): Promise<T | null> {
4282 | const headers = {
4283 | "User-Agent": USER_AGENT,
4284 | Accept: "application/geo+json",
4285 | };
4286 |
4287 | try {
4288 | const response = await fetch(url, { headers });
4289 | if (!response.ok) {
4290 | throw new Error(`HTTP error! status: ${response.status}`);
4291 | }
4292 | return (await response.json()) as T;
4293 | } catch (error) {
4294 | console.error("Error making NWS request:", error);
4295 | return null;
4296 | }
4297 | }
4298 |
4299 | interface AlertFeature {
4300 | properties: {
4301 | event?: string;
4302 | areaDesc?: string;
4303 | severity?: string;
4304 | status?: string;
4305 | headline?: string;
4306 | };
4307 | }
4308 |
4309 | // Format alert data
4310 | function formatAlert(feature: AlertFeature): string {
4311 | const props = feature.properties;
4312 | return [
4313 | `Event: ${props.event || "Unknown"}`,
4314 | `Area: ${props.areaDesc || "Unknown"}`,
4315 | `Severity: ${props.severity || "Unknown"}`,
4316 | `Status: ${props.status || "Unknown"}`,
4317 | `Headline: ${props.headline || "No headline"}`,
4318 | "---",
4319 | ].join("\n");
4320 | }
4321 |
4322 | interface ForecastPeriod {
4323 | name?: string;
4324 | temperature?: number;
4325 | temperatureUnit?: string;
4326 | windSpeed?: string;
4327 | windDirection?: string;
4328 | shortForecast?: string;
4329 | }
4330 |
4331 | interface AlertsResponse {
4332 | features: AlertFeature[];
4333 | }
4334 |
4335 | interface PointsResponse {
4336 | properties: {
4337 | forecast?: string;
4338 | };
4339 | }
4340 |
4341 | interface ForecastResponse {
4342 | properties: {
4343 | periods: ForecastPeriod[];
4344 | };
4345 | }
4346 | ```
4347 |
4348 | ### Implementing tool execution
4349 |
4350 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4351 |
4352 | ```typescript
4353 | // Register weather tools
4354 | server.tool(
4355 | "get-alerts",
4356 | "Get weather alerts for a state",
4357 | {
4358 | state: z.string().length(2).describe("Two-letter state code (e.g. CA, NY)"),
4359 | },
4360 | async ({ state }) => {
4361 | const stateCode = state.toUpperCase();
4362 | const alertsUrl = `${NWS_API_BASE}/alerts?area=${stateCode}`;
4363 | const alertsData = await makeNWSRequest<AlertsResponse>(alertsUrl);
4364 |
4365 | if (!alertsData) {
4366 | return {
4367 | content: [
4368 | {
4369 | type: "text",
4370 | text: "Failed to retrieve alerts data",
4371 | },
4372 | ],
4373 | };
4374 | }
4375 |
4376 | const features = alertsData.features || [];
4377 | if (features.length === 0) {
4378 | return {
4379 | content: [
4380 | {
4381 | type: "text",
4382 | text: `No active alerts for ${stateCode}`,
4383 | },
4384 | ],
4385 | };
4386 | }
4387 |
4388 | const formattedAlerts = features.map(formatAlert);
4389 | const alertsText = `Active alerts for ${stateCode}:\n\n${formattedAlerts.join("\n")}`;
4390 |
4391 | return {
4392 | content: [
4393 | {
4394 | type: "text",
4395 | text: alertsText,
4396 | },
4397 | ],
4398 | };
4399 | },
4400 | );
4401 |
4402 | server.tool(
4403 | "get-forecast",
4404 | "Get weather forecast for a location",
4405 | {
4406 | latitude: z.number().min(-90).max(90).describe("Latitude of the location"),
4407 | longitude: z.number().min(-180).max(180).describe("Longitude of the location"),
4408 | },
4409 | async ({ latitude, longitude }) => {
4410 | // Get grid point data
4411 | const pointsUrl = `${NWS_API_BASE}/points/${latitude.toFixed(4)},${longitude.toFixed(4)}`;
4412 | const pointsData = await makeNWSRequest<PointsResponse>(pointsUrl);
4413 |
4414 | if (!pointsData) {
4415 | return {
4416 | content: [
4417 | {
4418 | type: "text",
4419 | text: `Failed to retrieve grid point data for coordinates: ${latitude}, ${longitude}. This location may not be supported by the NWS API (only US locations are supported).`,
4420 | },
4421 | ],
4422 | };
4423 | }
4424 |
4425 | const forecastUrl = pointsData.properties?.forecast;
4426 | if (!forecastUrl) {
4427 | return {
4428 | content: [
4429 | {
4430 | type: "text",
4431 | text: "Failed to get forecast URL from grid point data",
4432 | },
4433 | ],
4434 | };
4435 | }
4436 |
4437 | // Get forecast data
4438 | const forecastData = await makeNWSRequest<ForecastResponse>(forecastUrl);
4439 | if (!forecastData) {
4440 | return {
4441 | content: [
4442 | {
4443 | type: "text",
4444 | text: "Failed to retrieve forecast data",
4445 | },
4446 | ],
4447 | };
4448 | }
4449 |
4450 | const periods = forecastData.properties?.periods || [];
4451 | if (periods.length === 0) {
4452 | return {
4453 | content: [
4454 | {
4455 | type: "text",
4456 | text: "No forecast periods available",
4457 | },
4458 | ],
4459 | };
4460 | }
4461 |
4462 | // Format forecast periods
4463 | const formattedForecast = periods.map((period: ForecastPeriod) =>
4464 | [
4465 | `${period.name || "Unknown"}:`,
4466 | `Temperature: ${period.temperature || "Unknown"}°${period.temperatureUnit || "F"}`,
4467 | `Wind: ${period.windSpeed || "Unknown"} ${period.windDirection || ""}`,
4468 | `${period.shortForecast || "No forecast available"}`,
4469 | "---",
4470 | ].join("\n"),
4471 | );
4472 |
4473 | const forecastText = `Forecast for ${latitude}, ${longitude}:\n\n${formattedForecast.join("\n")}`;
4474 |
4475 | return {
4476 | content: [
4477 | {
4478 | type: "text",
4479 | text: forecastText,
4480 | },
4481 | ],
4482 | };
4483 | },
4484 | );
4485 | ```
4486 |
4487 | ### Running the server
4488 |
4489 | Finally, implement the main function to run the server:
4490 |
4491 | ```typescript
4492 | async function main() {
4493 | const transport = new StdioServerTransport();
4494 | await server.connect(transport);
4495 | console.error("Weather MCP Server running on stdio");
4496 | }
4497 |
4498 | main().catch((error) => {
4499 | console.error("Fatal error in main():", error);
4500 | process.exit(1);
4501 | });
4502 | ```
4503 |
4504 | Make sure to run `npm run build` to build your server! This is a very important step in getting your server to connect.
4505 |
4506 | Let's now test your server from an existing MCP host, Claude for Desktop.
4507 |
4508 | ## Testing your server with Claude for Desktop
4509 |
4510 | <Note>
4511 | Claude for Desktop is not yet available on Linux. Linux users can proceed to the [Building a client](/quickstart/client) tutorial to build an MCP client that connects to the server we just built.
4512 | </Note>
4513 |
4514 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4515 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4516 |
4517 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use. To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor. Make sure to create the file if it doesn't exist.
4518 |
4519 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4520 |
4521 | <Tabs>
4522 | <Tab title="MacOS/Linux">
4523 | ```bash
4524 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4525 | ```
4526 | </Tab>
4527 |
4528 | <Tab title="Windows">
4529 | ```powershell
4530 | code $env:AppData\Claude\claude_desktop_config.json
4531 | ```
4532 | </Tab>
4533 | </Tabs>
4534 |
4535 | You'll then add your servers in the `mcpServers` key. The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4536 |
4537 | In this case, we'll add our single weather server like so:
4538 |
4539 | <Tabs>
4540 | <Tab title="MacOS/Linux">
4541 | <CodeGroup>
4542 | ```json Node
4543 | {
4544 | "mcpServers": {
4545 | "weather": {
4546 | "command": "node",
4547 | "args": [
4548 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js"
4549 | ]
4550 | }
4551 | }
4552 | }
4553 | ```
4554 | </CodeGroup>
4555 | </Tab>
4556 |
4557 | <Tab title="Windows">
4558 | <CodeGroup>
4559 | ```json Node
4560 | {
4561 | "mcpServers": {
4562 | "weather": {
4563 | "command": "node",
4564 | "args": [
4565 | "C:\\PATH\\TO\\PARENT\\FOLDER\\weather\\build\\index.js"
4566 | ]
4567 | }
4568 | }
4569 | }
4570 | ```
4571 | </CodeGroup>
4572 | </Tab>
4573 | </Tabs>
4574 |
4575 | This tells Claude for Desktop:
4576 |
4577 | 1. There's an MCP server named "weather"
4578 | 2. Launch it by running `node /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js`
4579 |
4580 | Save the file, and restart **Claude for Desktop**.
4581 | </Tab>
4582 |
4583 | <Tab title="Java">
4584 | <Note>
4585 | This is a quickstart demo based on Spring AI MCP auto-configuraiton and boot starters.
4586 | To learn how to create sync and async MCP Servers, manually, consult the [Java SDK Server](/sdk/java/mcp-server) documentation.
4587 | </Note>
4588 |
4589 | Let's get started with building our weather server!
4590 | [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)
4591 |
4592 | 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.
4593 | For manual MCP Server implementation, refer to the [MCP Server Java SDK documentation](/sdk/java/mcp-server).
4594 |
4595 | ### System requirements
4596 |
4597 | * Java 17 or higher installed.
4598 | * [Spring Boot 3.3.x](https://docs.spring.io/spring-boot/installing.html) or higher
4599 |
4600 | ### Set up your environment
4601 |
4602 | Use the [Spring Initizer](https://start.spring.io/) to bootstrat the project.
4603 |
4604 | You will need to add the following dependencies:
4605 |
4606 | <Tabs>
4607 | <Tab title="Maven">
4608 | ```xml
4609 | <dependencies>
4610 | <dependency>
4611 | <groupId>org.springframework.ai</groupId>
4612 | <artifactId>spring-ai-mcp-server-spring-boot-starter</artifactId>
4613 | </dependency>
4614 |
4615 | <dependency>
4616 | <groupId>org.springframework</groupId>
4617 | <artifactId>spring-web</artifactId>
4618 | </dependency>
4619 | </dependencies>
4620 | ```
4621 | </Tab>
4622 |
4623 | <Tab title="Gradle">
4624 | ```groovy
4625 | dependencies {
4626 | implementation platform("org.springframework.ai:spring-ai-mcp-server-spring-boot-starter")
4627 | implementation platform("org.springframework:spring-web")
4628 | }
4629 | ```
4630 | </Tab>
4631 | </Tabs>
4632 |
4633 | Then configure your application by setting the applicaiton properties:
4634 |
4635 | <CodeGroup>
4636 | ```bash application.properties
4637 | spring.main.bannerMode=off
4638 | logging.pattern.console=
4639 | ```
4640 |
4641 | ```yaml application.yml
4642 | logging:
4643 | pattern:
4644 | console:
4645 | spring:
4646 | main:
4647 | banner-mode: off
4648 | ```
4649 | </CodeGroup>
4650 |
4651 | 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.
4652 |
4653 | Now let's dive into building your server.
4654 |
4655 | ## Building your server
4656 |
4657 | ### Weather Service
4658 |
4659 | 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:
4660 |
4661 | ```java
4662 | @Service
4663 | public class WeatherService {
4664 |
4665 | private final RestClient restClient;
4666 |
4667 | public WeatherService() {
4668 | this.restClient = RestClient.builder()
4669 | .baseUrl("https://api.weather.gov")
4670 | .defaultHeader("Accept", "application/geo+json")
4671 | .defaultHeader("User-Agent", "WeatherApiClient/1.0 ([email protected])")
4672 | .build();
4673 | }
4674 |
4675 | @Tool(description = "Get weather forecast for a specific latitude/longitude")
4676 | public String getWeatherForecastByLocation(
4677 | double latitude, // Latitude coordinate
4678 | double longitude // Longitude coordinate
4679 | ) {
4680 | // Returns detailed forecast including:
4681 | // - Temperature and unit
4682 | // - Wind speed and direction
4683 | // - Detailed forecast description
4684 | }
4685 |
4686 | @Tool(description = "Get weather alerts for a US state")
4687 | public String getAlerts(
4688 | @ToolParam(description = "Two-letter US state code (e.g. CA, NY") String state)
4689 | ) {
4690 | // Returns active alerts including:
4691 | // - Event type
4692 | // - Affected area
4693 | // - Severity
4694 | // - Description
4695 | // - Safety instructions
4696 | }
4697 |
4698 | // ......
4699 | }
4700 | ```
4701 |
4702 | The `@Service` annotation with auto-register the service in your applicaiton context.
4703 | The Spring AI `@Tool` annotation, making it easy to create and maintain MCP tools.
4704 |
4705 | The auto-configuration will automatically register these tools with the MCP server.
4706 |
4707 | ### Create your Boot Applicaiton
4708 |
4709 | ```java
4710 | @SpringBootApplication
4711 | public class McpServerApplication {
4712 |
4713 | public static void main(String[] args) {
4714 | SpringApplication.run(McpServerApplication.class, args);
4715 | }
4716 |
4717 | @Bean
4718 | public ToolCallbackProvider weatherTools(WeatherService weatherService) {
4719 | return MethodToolCallbackProvider.builder().toolObjects(weatherService).build();
4720 | }
4721 | }
4722 | ```
4723 |
4724 | Uses the the `MethodToolCallbackProvider` utils to convert the `@Tools` into actionalble callbackes used by the MCP server.
4725 |
4726 | ### Running the server
4727 |
4728 | Finally, let's build the server:
4729 |
4730 | ```bash
4731 | ./mvnw clean install
4732 | ```
4733 |
4734 | This will generate a `mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar` file within the `target` folder.
4735 |
4736 | Let's now test your server from an existing MCP host, Claude for Desktop.
4737 |
4738 | ## Testing your server with Claude for Desktop
4739 |
4740 | <Note>
4741 | Claude for Desktop is not yet available on Linux.
4742 | </Note>
4743 |
4744 | First, make sure you have Claude for Desktop installed.
4745 | [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.**
4746 |
4747 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use.
4748 | To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor.
4749 | Make sure to create the file if it doesn't exist.
4750 |
4751 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4752 |
4753 | <Tabs>
4754 | <Tab title="MacOS/Linux">
4755 | ```bash
4756 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4757 | ```
4758 | </Tab>
4759 |
4760 | <Tab title="Windows">
4761 | ```powershell
4762 | code $env:AppData\Claude\claude_desktop_config.json
4763 | ```
4764 | </Tab>
4765 | </Tabs>
4766 |
4767 | You'll then add your servers in the `mcpServers` key.
4768 | The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4769 |
4770 | In this case, we'll add our single weather server like so:
4771 |
4772 | <Tabs>
4773 | <Tab title="MacOS/Linux">
4774 | ```json java
4775 | {
4776 | "mcpServers": {
4777 | "spring-ai-mcp-weather": {
4778 | "command": "java",
4779 | "args": [
4780 | "-Dspring.ai.mcp.server.stdio=true",
4781 | "-jar",
4782 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4783 | ]
4784 | }
4785 | }
4786 | }
4787 | ```
4788 | </Tab>
4789 |
4790 | <Tab title="Windows">
4791 | ```json java
4792 | {
4793 | "mcpServers": {
4794 | "spring-ai-mcp-weather": {
4795 | "command": "java",
4796 | "args": [
4797 | "-Dspring.ai.mcp.server.transport=STDIO",
4798 | "-jar",
4799 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather\\mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4800 | ]
4801 | }
4802 | }
4803 | }
4804 | ```
4805 | </Tab>
4806 | </Tabs>
4807 |
4808 | <Note>
4809 | Make sure you pass in the absolute path to your server.
4810 | </Note>
4811 |
4812 | This tells Claude for Desktop:
4813 |
4814 | 1. There's an MCP server named "my-weather-server"
4815 | 2. To launch it by running `java -jar /ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar`
4816 |
4817 | Save the file, and restart **Claude for Desktop**.
4818 |
4819 | ## Testing your server with Java client
4820 |
4821 | ### Create a MCP Client manually
4822 |
4823 | Use the `McpClient` to connect to the server:
4824 |
4825 | ```java
4826 | var stdioParams = ServerParameters.builder("java")
4827 | .args("-jar", "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar")
4828 | .build();
4829 |
4830 | var stdioTransport = new StdioClientTransport(stdioParams);
4831 |
4832 | var mcpClient = McpClient.sync(stdioTransport).build();
4833 |
4834 | mcpClient.initialize();
4835 |
4836 | ListToolsResult toolsList = mcpClient.listTools();
4837 |
4838 | CallToolResult weather = mcpClient.callTool(
4839 | new CallToolRequest("getWeatherForecastByLocation",
4840 | Map.of("latitude", "47.6062", "longitude", "-122.3321")));
4841 |
4842 | CallToolResult alert = mcpClient.callTool(
4843 | new CallToolRequest("getAlerts", Map.of("state", "NY")));
4844 |
4845 | mcpClient.closeGracefully();
4846 | ```
4847 |
4848 | ### Use MCP Client Boot Starter
4849 |
4850 | Create a new boot starter applicaiton using the `spring-ai-mcp-client-spring-boot-starter` dependency:
4851 |
4852 | ```xml
4853 | <dependency>
4854 | <groupId>org.springframework.ai</groupId>
4855 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
4856 | </dependency>
4857 | ```
4858 |
4859 | and set the `spring.ai.mcp.client.stdio.servers-configuration` property to point to your `claude_desktop_config.json`.
4860 | You can re-use the existing Anthropic Destop configuration:
4861 |
4862 | ```properties
4863 | spring.ai.mcp.client.stdio.servers-configuration=file:PATH/TO/claude_desktop_config.json
4864 | ```
4865 |
4866 | When you stasrt your client applicaiton, the auto-configuration will create, automatically MCP clients from the claude\_desktop\_config.json.
4867 |
4868 | 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.
4869 |
4870 | ## More Java MCP Server examples
4871 |
4872 | 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.
4873 | It showcases how to define and register MCP Tools, Resources, and Prompts, using the Spring Boot's auto-configuration capabilities.
4874 | </Tab>
4875 | </Tabs>
4876 |
4877 | ### Test with commands
4878 |
4879 | 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:
4880 |
4881 | <Frame>
4882 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/visual-indicator-mcp-tools.png" />
4883 | </Frame>
4884 |
4885 | After clicking on the hammer icon, you should see two tools listed:
4886 |
4887 | <Frame>
4888 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/available-mcp-tools.png" />
4889 | </Frame>
4890 |
4891 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
4892 |
4893 | If the hammer icon has shown up, you can now test your server by running the following commands in Claude for Desktop:
4894 |
4895 | * What's the weather in Sacramento?
4896 | * What are the active weather alerts in Texas?
4897 |
4898 | <Frame>
4899 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
4900 | </Frame>
4901 |
4902 | <Frame>
4903 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
4904 | </Frame>
4905 |
4906 | <Note>
4907 | Since this is the US National Weather service, the queries will only work for US locations.
4908 | </Note>
4909 |
4910 | ## What's happening under the hood
4911 |
4912 | When you ask a question:
4913 |
4914 | 1. The client sends your question to Claude
4915 | 2. Claude analyzes the available tools and decides which one(s) to use
4916 | 3. The client executes the chosen tool(s) through the MCP server
4917 | 4. The results are sent back to Claude
4918 | 5. Claude formulates a natural language response
4919 | 6. The response is displayed to you!
4920 |
4921 | ## Troubleshooting
4922 |
4923 | <AccordionGroup>
4924 | <Accordion title="Claude for Desktop Integration Issues">
4925 | **Getting logs from Claude for Desktop**
4926 |
4927 | Claude.app logging related to MCP is written to log files in `~/Library/Logs/Claude`:
4928 |
4929 | * `mcp.log` will contain general logging about MCP connections and connection failures.
4930 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
4931 |
4932 | You can run the following command to list recent logs and follow along with any new ones:
4933 |
4934 | ```bash
4935 | # Check Claude's logs for errors
4936 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
4937 | ```
4938 |
4939 | **Server not showing up in Claude**
4940 |
4941 | 1. Check your `claude_desktop_config.json` file syntax
4942 | 2. Make sure the path to your project is absolute and not relative
4943 | 3. Restart Claude for Desktop completely
4944 |
4945 | **Tool calls failing silently**
4946 |
4947 | If Claude attempts to use the tools but they fail:
4948 |
4949 | 1. Check Claude's logs for errors
4950 | 2. Verify your server builds and runs without errors
4951 | 3. Try restarting Claude for Desktop
4952 |
4953 | **None of this is working. What do I do?**
4954 |
4955 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
4956 | </Accordion>
4957 |
4958 | <Accordion title="Weather API Issues">
4959 | **Error: Failed to retrieve grid point data**
4960 |
4961 | This usually means either:
4962 |
4963 | 1. The coordinates are outside the US
4964 | 2. The NWS API is having issues
4965 | 3. You're being rate limited
4966 |
4967 | Fix:
4968 |
4969 | * Verify you're using US coordinates
4970 | * Add a small delay between requests
4971 | * Check the NWS API status page
4972 |
4973 | **Error: No active alerts for \[STATE]**
4974 |
4975 | 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.
4976 | </Accordion>
4977 | </AccordionGroup>
4978 |
4979 | <Note>
4980 | For more advanced troubleshooting, check out our guide on [Debugging MCP](/docs/tools/debugging)
4981 | </Note>
4982 |
4983 | ## Next steps
4984 |
4985 | <CardGroup cols={2}>
4986 | <Card title="Building a client" icon="outlet" href="/quickstart/client">
4987 | Learn how to build your own MCP client that can connect to your server
4988 | </Card>
4989 |
4990 | <Card title="Example servers" icon="grid" href="/examples">
4991 | Check out our gallery of official MCP servers and implementations
4992 | </Card>
4993 |
4994 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
4995 | Learn how to effectively debug MCP servers and integrations
4996 | </Card>
4997 |
4998 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
4999 | Learn how to use LLMs like Claude to speed up your MCP development
5000 | </Card>
5001 | </CardGroup>
5002 |
5003 |
5004 | # For Claude Desktop Users
5005 | Source: https://modelcontextprotocol.io/quickstart/user
5006 |
5007 | Get started using pre-built servers in Claude for Desktop.
5008 |
5009 | 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.
5010 |
5011 | <Frame>
5012 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-filesystem.png" />
5013 | </Frame>
5014 |
5015 | Don't worry — it will ask you for your permission before executing these actions!
5016 |
5017 | ## 1. Download Claude for Desktop
5018 |
5019 | Start by downloading [Claude for Desktop](https://claude.ai/download), choosing either macOS or Windows. (Linux is not yet supported for Claude for Desktop.)
5020 |
5021 | Follow the installation instructions.
5022 |
5023 | 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..."
5024 |
5025 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
5026 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
5027 | </Accordion>
5028 |
5029 | ## 2. Add the Filesystem MCP Server
5030 |
5031 | 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.
5032 |
5033 | 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.
5034 |
5035 | This is what it should look like on a Mac:
5036 |
5037 | <Frame style={{ textAlign: 'center' }}>
5038 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-menu.png" width="400" />
5039 | </Frame>
5040 |
5041 | Click on "Developer" in the lefthand bar of the Settings pane, and then click on "Edit Config":
5042 |
5043 | <Frame>
5044 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-developer.png" />
5045 | </Frame>
5046 |
5047 | This will create a configuration file at:
5048 |
5049 | * macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
5050 | * Windows: `%APPDATA%\Claude\claude_desktop_config.json`
5051 |
5052 | if you don't already have one, and will display the file in your file system.
5053 |
5054 | Open up the configuration file in any text editor. Replace the file contents with this:
5055 |
5056 | <Tabs>
5057 | <Tab title="MacOS/Linux">
5058 | ```json
5059 | {
5060 | "mcpServers": {
5061 | "filesystem": {
5062 | "command": "npx",
5063 | "args": [
5064 | "-y",
5065 | "@modelcontextprotocol/server-filesystem",
5066 | "/Users/username/Desktop",
5067 | "/Users/username/Downloads"
5068 | ]
5069 | }
5070 | }
5071 | }
5072 | ```
5073 | </Tab>
5074 |
5075 | <Tab title="Windows">
5076 | ```json
5077 | {
5078 | "mcpServers": {
5079 | "filesystem": {
5080 | "command": "npx",
5081 | "args": [
5082 | "-y",
5083 | "@modelcontextprotocol/server-filesystem",
5084 | "C:\\Users\\username\\Desktop",
5085 | "C:\\Users\\username\\Downloads"
5086 | ]
5087 | }
5088 | }
5089 | }
5090 | ```
5091 | </Tab>
5092 | </Tabs>
5093 |
5094 | 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.
5095 |
5096 | 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.
5097 |
5098 | * On macOS, open the Terminal from your Applications folder
5099 | * On Windows, press Windows + R, type "cmd", and press Enter
5100 |
5101 | Once in the command line, verify you have Node installed by entering in the following command:
5102 |
5103 | ```bash
5104 | node --version
5105 | ```
5106 |
5107 | If you get an error saying "command not found" or "node is not recognized", download Node from [nodejs.org](https://nodejs.org/).
5108 |
5109 | <Tip>
5110 | **How does the configuration file work?**
5111 |
5112 | 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.
5113 | </Tip>
5114 |
5115 | <Warning>
5116 | **Command Privileges**
5117 |
5118 | 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.
5119 | </Warning>
5120 |
5121 | ## 3. Restart Claude
5122 |
5123 | After updating your configuration file, you need to restart Claude for Desktop.
5124 |
5125 | 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:
5126 |
5127 | <Frame>
5128 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-hammer.png" />
5129 | </Frame>
5130 |
5131 | After clicking on the hammer icon, you should see the tools that come with the Filesystem MCP Server:
5132 |
5133 | <Frame style={{ textAlign: 'center' }}>
5134 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-tools.png" width="400" />
5135 | </Frame>
5136 |
5137 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
5138 |
5139 | ## 4. Try it out!
5140 |
5141 | You can now talk to Claude and ask it about your filesystem. It should know when to call the relevant tools.
5142 |
5143 | Things you might try asking Claude:
5144 |
5145 | * Can you write a poem and save it to my desktop?
5146 | * What are some work-related files in my downloads folder?
5147 | * Can you take all the images on my desktop and move them to a new folder called "Images"?
5148 |
5149 | As needed, Claude will call the relevant tools and seek your approval before taking an action:
5150 |
5151 | <Frame style={{ textAlign: 'center' }}>
5152 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-approve.png" width="500" />
5153 | </Frame>
5154 |
5155 | ## Troubleshooting
5156 |
5157 | <AccordionGroup>
5158 | <Accordion title="Server not showing up in Claude / hammer icon missing">
5159 | 1. Restart Claude for Desktop completely
5160 | 2. Check your `claude_desktop_config.json` file syntax
5161 | 3. Make sure the file paths included in `claude_desktop_config.json` are valid and that they are absolute and not relative
5162 | 4. Look at [logs](#getting-logs-from-claude-for-desktop) to see why the server is not connecting
5163 | 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:
5164 |
5165 | <Tabs>
5166 | <Tab title="MacOS/Linux">
5167 | ```bash
5168 | npx -y @modelcontextprotocol/server-filesystem /Users/username/Desktop /Users/username/Downloads
5169 | ```
5170 | </Tab>
5171 |
5172 | <Tab title="Windows">
5173 | ```bash
5174 | npx -y @modelcontextprotocol/server-filesystem C:\Users\username\Desktop C:\Users\username\Downloads
5175 | ```
5176 | </Tab>
5177 | </Tabs>
5178 | </Accordion>
5179 |
5180 | <Accordion title="Getting logs from Claude for Desktop">
5181 | Claude.app logging related to MCP is written to log files in:
5182 |
5183 | * macOS: `~/Library/Logs/Claude`
5184 |
5185 | * Windows: `%APPDATA%\Claude\logs`
5186 |
5187 | * `mcp.log` will contain general logging about MCP connections and connection failures.
5188 |
5189 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
5190 |
5191 | 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):
5192 |
5193 | <Tabs>
5194 | <Tab title="MacOS/Linux">
5195 | ```bash
5196 | # Check Claude's logs for errors
5197 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
5198 | ```
5199 | </Tab>
5200 |
5201 | <Tab title="Windows">
5202 | ```bash
5203 | type "%APPDATA%\Claude\logs\mcp*.log"
5204 | ```
5205 | </Tab>
5206 | </Tabs>
5207 | </Accordion>
5208 |
5209 | <Accordion title="Tool calls failing silently">
5210 | If Claude attempts to use the tools but they fail:
5211 |
5212 | 1. Check Claude's logs for errors
5213 | 2. Verify your server builds and runs without errors
5214 | 3. Try restarting Claude for Desktop
5215 | </Accordion>
5216 |
5217 | <Accordion title="None of this is working. What do I do?">
5218 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
5219 | </Accordion>
5220 |
5221 | <Accordion title="ENOENT error and `${APPDATA}` in paths on Windows">
5222 | If your configured server fails to load, and you see within its logs an error referring to `${APPDATA}` within a path, you may need to add the expanded value of `%APPDATA%` to your `env` key in `claude_desktop_config.json`:
5223 |
5224 | ```json
5225 | {
5226 | "brave-search": {
5227 | "command": "npx",
5228 | "args": ["-y", "@modelcontextprotocol/server-brave-search"],
5229 | "env": {
5230 | "APPDATA": "C:\\Users\\user\\AppData\\Roaming\\",
5231 | "BRAVE_API_KEY": "..."
5232 | }
5233 | }
5234 | }
5235 | ```
5236 |
5237 | With this change in place, launch Claude Desktop once again.
5238 |
5239 | <Warning>
5240 | **NPM should be installed globally**
5241 |
5242 | 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:
5243 |
5244 | ```bash
5245 | npm install -g npm
5246 | ```
5247 | </Warning>
5248 | </Accordion>
5249 | </AccordionGroup>
5250 |
5251 | ## Next steps
5252 |
5253 | <CardGroup cols={2}>
5254 | <Card title="Explore other servers" icon="grid" href="/examples">
5255 | Check out our gallery of official MCP servers and implementations
5256 | </Card>
5257 |
5258 | <Card title="Build your own server" icon="code" href="/quickstart/server">
5259 | Now build your own custom server to use in Claude for Desktop and other clients
5260 | </Card>
5261 | </CardGroup>
5262 |
5263 |
5264 | # MCP Client
5265 | Source: https://modelcontextprotocol.io/sdk/java/mcp-client
5266 |
5267 | Learn how to use the Model Context Protocol (MCP) client to interact with MCP servers
5268 |
5269 | # Model Context Protocol Client
5270 |
5271 | 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:
5272 |
5273 | * Protocol version negotiation to ensure compatibility with servers
5274 | * Capability negotiation to determine available features
5275 | * Message transport and JSON-RPC communication
5276 | * Tool discovery and execution
5277 | * Resource access and management
5278 | * Prompt system interactions
5279 | * Optional features like roots management and sampling support
5280 |
5281 | The client provides both synchronous and asynchronous APIs for flexibility in different application contexts.
5282 |
5283 | <Tabs>
5284 | <Tab title="Sync API">
5285 | ```java
5286 | // Create a sync client with custom configuration
5287 | McpSyncClient client = McpClient.sync(transport)
5288 | .requestTimeout(Duration.ofSeconds(10))
5289 | .capabilities(ClientCapabilities.builder()
5290 | .roots(true) // Enable roots capability
5291 | .sampling() // Enable sampling capability
5292 | .build())
5293 | .sampling(request -> new CreateMessageResult(response))
5294 | .build();
5295 |
5296 | // Initialize connection
5297 | client.initialize();
5298 |
5299 | // List available tools
5300 | ListToolsResult tools = client.listTools();
5301 |
5302 | // Call a tool
5303 | CallToolResult result = client.callTool(
5304 | new CallToolRequest("calculator",
5305 | Map.of("operation", "add", "a", 2, "b", 3))
5306 | );
5307 |
5308 | // List and read resources
5309 | ListResourcesResult resources = client.listResources();
5310 | ReadResourceResult resource = client.readResource(
5311 | new ReadResourceRequest("resource://uri")
5312 | );
5313 |
5314 | // List and use prompts
5315 | ListPromptsResult prompts = client.listPrompts();
5316 | GetPromptResult prompt = client.getPrompt(
5317 | new GetPromptRequest("greeting", Map.of("name", "Spring"))
5318 | );
5319 |
5320 | // Add/remove roots
5321 | client.addRoot(new Root("file:///path", "description"));
5322 | client.removeRoot("file:///path");
5323 |
5324 | // Close client
5325 | client.closeGracefully();
5326 | ```
5327 | </Tab>
5328 |
5329 | <Tab title="Async API">
5330 | ```java
5331 | // Create an async client with custom configuration
5332 | McpAsyncClient client = McpClient.async(transport)
5333 | .requestTimeout(Duration.ofSeconds(10))
5334 | .capabilities(ClientCapabilities.builder()
5335 | .roots(true) // Enable roots capability
5336 | .sampling() // Enable sampling capability
5337 | .build())
5338 | .sampling(request -> Mono.just(new CreateMessageResult(response)))
5339 | .toolsChangeConsumer(tools -> Mono.fromRunnable(() -> {
5340 | logger.info("Tools updated: {}", tools);
5341 | }))
5342 | .resourcesChangeConsumer(resources -> Mono.fromRunnable(() -> {
5343 | logger.info("Resources updated: {}", resources);
5344 | }))
5345 | .promptsChangeConsumer(prompts -> Mono.fromRunnable(() -> {
5346 | logger.info("Prompts updated: {}", prompts);
5347 | }))
5348 | .build();
5349 |
5350 | // Initialize connection and use features
5351 | client.initialize()
5352 | .flatMap(initResult -> client.listTools())
5353 | .flatMap(tools -> {
5354 | return client.callTool(new CallToolRequest(
5355 | "calculator",
5356 | Map.of("operation", "add", "a", 2, "b", 3)
5357 | ));
5358 | })
5359 | .flatMap(result -> {
5360 | return client.listResources()
5361 | .flatMap(resources ->
5362 | client.readResource(new ReadResourceRequest("resource://uri"))
5363 | );
5364 | })
5365 | .flatMap(resource -> {
5366 | return client.listPrompts()
5367 | .flatMap(prompts ->
5368 | client.getPrompt(new GetPromptRequest(
5369 | "greeting",
5370 | Map.of("name", "Spring")
5371 | ))
5372 | );
5373 | })
5374 | .flatMap(prompt -> {
5375 | return client.addRoot(new Root("file:///path", "description"))
5376 | .then(client.removeRoot("file:///path"));
5377 | })
5378 | .doFinally(signalType -> {
5379 | client.closeGracefully().subscribe();
5380 | })
5381 | .subscribe();
5382 | ```
5383 | </Tab>
5384 | </Tabs>
5385 |
5386 | ## Client Transport
5387 |
5388 | 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.
5389 |
5390 | <Tabs>
5391 | <Tab title="STDIO">
5392 | Creates transport for in-process based communication
5393 |
5394 | ```java
5395 | ServerParameters params = ServerParameters.builder("npx")
5396 | .args("-y", "@modelcontextprotocol/server-everything", "dir")
5397 | .build();
5398 | McpTransport transport = new StdioClientTransport(params);
5399 | ```
5400 | </Tab>
5401 |
5402 | <Tab title="SSE (HttpClient)">
5403 | Creates a framework agnostic (pure Java API) SSE client transport. Included in the core mcp module.
5404 |
5405 | ```java
5406 | McpTransport transport = new HttpClientSseClientTransport("http://your-mcp-server");
5407 | ```
5408 | </Tab>
5409 |
5410 | <Tab title="SSE (WebFlux)">
5411 | Creates WebFlux-based SSE client transport. Requires the mcp-webflux-sse-transport dependency.
5412 |
5413 | ```java
5414 | WebClient.Builder webClientBuilder = WebClient.builder()
5415 | .baseUrl("http://your-mcp-server");
5416 | McpTransport transport = new WebFluxSseClientTransport(webClientBuilder);
5417 | ```
5418 | </Tab>
5419 | </Tabs>
5420 |
5421 | ## Client Capabilities
5422 |
5423 | The client can be configured with various capabilities:
5424 |
5425 | ```java
5426 | var capabilities = ClientCapabilities.builder()
5427 | .roots(true) // Enable filesystem roots support with list changes notifications
5428 | .sampling() // Enable LLM sampling support
5429 | .build();
5430 | ```
5431 |
5432 | ### Roots Support
5433 |
5434 | Roots define the boundaries of where servers can operate within the filesystem:
5435 |
5436 | ```java
5437 | // Add a root dynamically
5438 | client.addRoot(new Root("file:///path", "description"));
5439 |
5440 | // Remove a root
5441 | client.removeRoot("file:///path");
5442 |
5443 | // Notify server of roots changes
5444 | client.rootsListChangedNotification();
5445 | ```
5446 |
5447 | The roots capability allows servers to:
5448 |
5449 | * Request the list of accessible filesystem roots
5450 | * Receive notifications when the roots list changes
5451 | * Understand which directories and files they have access to
5452 |
5453 | ### Sampling Support
5454 |
5455 | Sampling enables servers to request LLM interactions ("completions" or "generations") through the client:
5456 |
5457 | ```java
5458 | // Configure sampling handler
5459 | Function<CreateMessageRequest, CreateMessageResult> samplingHandler = request -> {
5460 | // Sampling implementation that interfaces with LLM
5461 | return new CreateMessageResult(response);
5462 | };
5463 |
5464 | // Create client with sampling support
5465 | var client = McpClient.sync(transport)
5466 | .capabilities(ClientCapabilities.builder()
5467 | .sampling()
5468 | .build())
5469 | .sampling(samplingHandler)
5470 | .build();
5471 | ```
5472 |
5473 | This capability allows:
5474 |
5475 | * Servers to leverage AI capabilities without requiring API keys
5476 | * Clients to maintain control over model access and permissions
5477 | * Support for both text and image-based interactions
5478 | * Optional inclusion of MCP server context in prompts
5479 |
5480 | ## Using MCP Clients
5481 |
5482 | ### Tool Execution
5483 |
5484 | 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.
5485 |
5486 | <Tabs>
5487 | <Tab title="Sync API">
5488 | ```java
5489 | // List available tools and their names
5490 | var tools = client.listTools();
5491 | tools.forEach(tool -> System.out.println(tool.getName()));
5492 |
5493 | // Execute a tool with parameters
5494 | var result = client.callTool("calculator", Map.of(
5495 | "operation", "add",
5496 | "a", 1,
5497 | "b", 2
5498 | ));
5499 | ```
5500 | </Tab>
5501 |
5502 | <Tab title="Async API">
5503 | ```java
5504 | // List available tools asynchronously
5505 | client.listTools()
5506 | .doOnNext(tools -> tools.forEach(tool ->
5507 | System.out.println(tool.getName())))
5508 | .subscribe();
5509 |
5510 | // Execute a tool asynchronously
5511 | client.callTool("calculator", Map.of(
5512 | "operation", "add",
5513 | "a", 1,
5514 | "b", 2
5515 | ))
5516 | .subscribe();
5517 | ```
5518 | </Tab>
5519 | </Tabs>
5520 |
5521 | ### Resource Access
5522 |
5523 | 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.
5524 |
5525 | <Tabs>
5526 | <Tab title="Sync API">
5527 | ```java
5528 | // List available resources and their names
5529 | var resources = client.listResources();
5530 | resources.forEach(resource -> System.out.println(resource.getName()));
5531 |
5532 | // Retrieve resource content using a URI template
5533 | var content = client.getResource("file", Map.of(
5534 | "path", "/path/to/file.txt"
5535 | ));
5536 | ```
5537 | </Tab>
5538 |
5539 | <Tab title="Async API">
5540 | ```java
5541 | // List available resources asynchronously
5542 | client.listResources()
5543 | .doOnNext(resources -> resources.forEach(resource ->
5544 | System.out.println(resource.getName())))
5545 | .subscribe();
5546 |
5547 | // Retrieve resource content asynchronously
5548 | client.getResource("file", Map.of(
5549 | "path", "/path/to/file.txt"
5550 | ))
5551 | .subscribe();
5552 | ```
5553 | </Tab>
5554 | </Tabs>
5555 |
5556 | ### Prompt System
5557 |
5558 | 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.
5559 |
5560 | <Tabs>
5561 | <Tab title="Sync API">
5562 | ```java
5563 | // List available prompt templates
5564 | var prompts = client.listPrompts();
5565 | prompts.forEach(prompt -> System.out.println(prompt.getName()));
5566 |
5567 | // Execute a prompt template with parameters
5568 | var response = client.executePrompt("echo", Map.of(
5569 | "text", "Hello, World!"
5570 | ));
5571 | ```
5572 | </Tab>
5573 |
5574 | <Tab title="Async API">
5575 | ```java
5576 | // List available prompt templates asynchronously
5577 | client.listPrompts()
5578 | .doOnNext(prompts -> prompts.forEach(prompt ->
5579 | System.out.println(prompt.getName())))
5580 | .subscribe();
5581 |
5582 | // Execute a prompt template asynchronously
5583 | client.executePrompt("echo", Map.of(
5584 | "text", "Hello, World!"
5585 | ))
5586 | .subscribe();
5587 | ```
5588 | </Tab>
5589 | </Tabs>
5590 |
5591 |
5592 | # Overview
5593 | Source: https://modelcontextprotocol.io/sdk/java/mcp-overview
5594 |
5595 | Introduction to the Model Context Protocol (MCP) Java SDK
5596 |
5597 | Java SDK for the [Model Context Protocol](https://modelcontextprotocol.org/docs/concepts/architecture)
5598 | enables standardized integration between AI models and tools.
5599 |
5600 | ## Features
5601 |
5602 | * MCP Client and MCP Server implementations supporting:
5603 | * Protocol [version compatibility negotiation](https://spec.modelcontextprotocol.io/specification/2024-11-05/basic/lifecycle/#initialization)
5604 | * [Tool](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/tools/) discovery, execution, list change notifications
5605 | * [Resource](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/resources/) management with URI templates
5606 | * [Roots](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/roots/) list management and notifications
5607 | * [Prompt](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/prompts/) handling and management
5608 | * [Sampling](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/sampling/) support for AI model interactions
5609 | * Multiple transport implementations:
5610 | * Default transports:
5611 | * Stdio-based transport for process-based communication
5612 | * Java HttpClient-based SSE client transport for HTTP SSE Client-side streaming
5613 | * Servlet-based SSE server transport for HTTP SSE Server streaming
5614 | * Spring-based transports:
5615 | * WebFlux SSE client and server transports for reactive HTTP streaming
5616 | * WebMVC SSE transport for servlet-based HTTP streaming
5617 | * Supports Synchronous and Asynchronous programming paradigms
5618 |
5619 | ## Architecture
5620 |
5621 | The SDK follows a layered architecture with clear separation of concerns:
5622 |
5623 | 
5624 |
5625 | * **Client/Server Layer (McpClient/McpServer)**: Both use McpSession for sync/async operations,
5626 | with McpClient handling client-side protocol operations and McpServer managing server-side protocol operations.
5627 | * **Session Layer (McpSession)**: Manages communication patterns and state using DefaultMcpSession implementation.
5628 | * **Transport Layer (McpTransport)**: Handles JSON-RPC message serialization/deserialization via:
5629 | * StdioTransport (stdin/stdout) in the core module
5630 | * HTTP SSE transports in dedicated transport modules (Java HttpClient, Spring WebFlux, Spring WebMVC)
5631 |
5632 | The MCP Client is a key component in the Model Context Protocol (MCP) architecture, responsible for establishing and managing connections with MCP servers.
5633 | It implements the client-side of the protocol.
5634 |
5635 | 
5636 |
5637 | The MCP Server is a foundational component in the Model Context Protocol (MCP) architecture that provides tools, resources, and capabilities to clients.
5638 | It implements the server-side of the protocol.
5639 |
5640 | 
5641 |
5642 | Key Interactions:
5643 |
5644 | * **Client/Server Initialization**: Transport setup, protocol compatibility check, capability negotiation, and implementation details exchange.
5645 | * **Message Flow**: JSON-RPC message handling with validation, type-safe response processing, and error handling.
5646 | * **Resource Management**: Resource discovery, URI template-based access, subscription system, and content retrieval.
5647 |
5648 | ## Dependencies
5649 |
5650 | Add the following Maven dependency to your project:
5651 |
5652 | <Tabs>
5653 | <Tab title="Maven">
5654 | The core MCP functionality:
5655 |
5656 | ```xml
5657 | <dependency>
5658 | <groupId>io.modelcontextprotocol.sdk</groupId>
5659 | <artifactId>mcp</artifactId>
5660 | </dependency>
5661 | ```
5662 |
5663 | For HTTP SSE transport implementations, add one of the following dependencies:
5664 |
5665 | ```xml
5666 | <!-- Spring WebFlux-based SSE client and server transport -->
5667 | <dependency>
5668 | <groupId>io.modelcontextprotocol.sdk</groupId>
5669 | <artifactId>mcp-spring-webflux</artifactId>
5670 | </dependency>
5671 |
5672 | <!-- Spring WebMVC-based SSE server transport -->
5673 | <dependency>
5674 | <groupId>io.modelcontextprotocol.sdk</groupId>
5675 | <artifactId>mcp-spring-webmvc</artifactId>
5676 | </dependency>
5677 | ```
5678 | </Tab>
5679 |
5680 | <Tab title="Gradle">
5681 | The core MCP functionality:
5682 |
5683 | ```groovy
5684 | dependencies {
5685 | implementation platform("io.modelcontextprotocol.sdk:mcp")
5686 | //...
5687 | }
5688 | ```
5689 |
5690 | For HTTP SSE transport implementations, add one of the following dependencies:
5691 |
5692 | ```groovy
5693 | // Spring WebFlux-based SSE client and server transport
5694 | dependencies {
5695 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webflux")
5696 | }
5697 |
5698 | // Spring WebMVC-based SSE server transport
5699 | dependencies {
5700 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webmvc")
5701 | }
5702 | ```
5703 | </Tab>
5704 | </Tabs>
5705 |
5706 | ### Bill of Materials (BOM)
5707 |
5708 | The Bill of Materials (BOM) declares the recommended versions of all the dependencies used by a given release.
5709 | Using the BOM from your application's build script avoids the need for you to specify and maintain the dependency versions yourself.
5710 | Instead, the version of the BOM you're using determines the utilized dependency versions.
5711 | It also ensures that you're using supported and tested versions of the dependencies by default, unless you choose to override them.
5712 |
5713 | Add the BOM to your project:
5714 |
5715 | <Tabs>
5716 | <Tab title="Maven">
5717 | ```xml
5718 | <dependencyManagement>
5719 | <dependencies>
5720 | <dependency>
5721 | <groupId>io.modelcontextprotocol.sdk</groupId>
5722 | <artifactId>mcp-bom</artifactId>
5723 | <version>0.7.0</version>
5724 | <type>pom</type>
5725 | <scope>import</scope>
5726 | </dependency>
5727 | </dependencies>
5728 | </dependencyManagement>
5729 | ```
5730 | </Tab>
5731 |
5732 | <Tab title="Gradle">
5733 | ```groovy
5734 | dependencies {
5735 | implementation platform("io.modelcontextprotocol.sdk:mcp-bom:0.7.0")
5736 | //...
5737 | }
5738 | ```
5739 |
5740 | 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.
5741 | This is implemented by adding a 'platform' dependency handler method to the dependencies section of your Gradle build script.
5742 | 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.
5743 | </Tab>
5744 | </Tabs>
5745 |
5746 | Replace the version number with the version of the BOM you want to use.
5747 |
5748 | ### Available Dependencies
5749 |
5750 | The following dependencies are available and managed by the BOM:
5751 |
5752 | * Core Dependencies
5753 | * `io.modelcontextprotocol.sdk:mcp` - Core MCP library providing the base functionality and APIs for Model Context Protocol implementation.
5754 | * Transport Dependencies
5755 | * `io.modelcontextprotocol.sdk:mcp-spring-webflux` - WebFlux-based Server-Sent Events (SSE) transport implementation for reactive applications.
5756 | * `io.modelcontextprotocol.sdk:mcp-spring-webmvc` - WebMVC-based Server-Sent Events (SSE) transport implementation for servlet-based applications.
5757 | * Testing Dependencies
5758 | * `io.modelcontextprotocol.sdk:mcp-test` - Testing utilities and support for MCP-based applications.
5759 |
5760 |
5761 | # MCP Server
5762 | Source: https://modelcontextprotocol.io/sdk/java/mcp-server
5763 |
5764 | Learn how to implement and configure a Model Context Protocol (MCP) server
5765 |
5766 | ## Overview
5767 |
5768 | 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:
5769 |
5770 | * Exposing tools that clients can discover and execute
5771 | * Managing resources with URI-based access patterns
5772 | * Providing prompt templates and handling prompt requests
5773 | * Supporting capability negotiation with clients
5774 | * Implementing server-side protocol operations
5775 | * Managing concurrent client connections
5776 | * Providing structured logging and notifications
5777 |
5778 | The server supports both synchronous and asynchronous APIs, allowing for flexible integration in different application contexts.
5779 |
5780 | <Tabs>
5781 | <Tab title="Sync API">
5782 | ```java
5783 | // Create a server with custom configuration
5784 | McpSyncServer syncServer = McpServer.sync(transport)
5785 | .serverInfo("my-server", "1.0.0")
5786 | .capabilities(ServerCapabilities.builder()
5787 | .resources(true) // Enable resource support
5788 | .tools(true) // Enable tool support
5789 | .prompts(true) // Enable prompt support
5790 | .logging() // Enable logging support
5791 | .build())
5792 | .build();
5793 |
5794 | // Register tools, resources, and prompts
5795 | syncServer.addTool(syncToolRegistration);
5796 | syncServer.addResource(syncResourceRegistration);
5797 | syncServer.addPrompt(syncPromptRegistration);
5798 |
5799 | // Send logging notifications
5800 | syncServer.loggingNotification(LoggingMessageNotification.builder()
5801 | .level(LoggingLevel.INFO)
5802 | .logger("custom-logger")
5803 | .data("Server initialized")
5804 | .build());
5805 |
5806 | // Close the server when done
5807 | syncServer.close();
5808 | ```
5809 | </Tab>
5810 |
5811 | <Tab title="Async API">
5812 | ```java
5813 | // Create an async server with custom configuration
5814 | McpAsyncServer asyncServer = McpServer.async(transport)
5815 | .serverInfo("my-server", "1.0.0")
5816 | .capabilities(ServerCapabilities.builder()
5817 | .resources(true) // Enable resource support
5818 | .tools(true) // Enable tool support
5819 | .prompts(true) // Enable prompt support
5820 | .logging() // Enable logging support
5821 | .build())
5822 | .build();
5823 |
5824 | // Register tools, resources, and prompts
5825 | asyncServer.addTool(asyncToolRegistration)
5826 | .doOnSuccess(v -> logger.info("Tool registered"))
5827 | .subscribe();
5828 |
5829 | asyncServer.addResource(asyncResourceRegistration)
5830 | .doOnSuccess(v -> logger.info("Resource registered"))
5831 | .subscribe();
5832 |
5833 | asyncServer.addPrompt(asyncPromptRegistration)
5834 | .doOnSuccess(v -> logger.info("Prompt registered"))
5835 | .subscribe();
5836 |
5837 | // Send logging notifications
5838 | asyncServer.loggingNotification(LoggingMessageNotification.builder()
5839 | .level(LoggingLevel.INFO)
5840 | .logger("custom-logger")
5841 | .data("Server initialized")
5842 | .build());
5843 |
5844 | // Close the server when done
5845 | asyncServer.close()
5846 | .doOnSuccess(v -> logger.info("Server closed"))
5847 | .subscribe();
5848 | ```
5849 | </Tab>
5850 | </Tabs>
5851 |
5852 | ## Server Transport
5853 |
5854 | 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:
5855 |
5856 | <Tabs>
5857 | <Tab title="STDIO">
5858 | <>
5859 | Create in-process based transport:
5860 |
5861 | ```java
5862 | StdioServerTransport transport = new StdioServerTransport(new ObjectMapper());
5863 | ```
5864 |
5865 | Provides bidirectional JSON-RPC message handling over standard input/output streams with non-blocking message processing, serialization/deserialization, and graceful shutdown support.
5866 |
5867 | Key features:
5868 |
5869 | <ul>
5870 | <li>Bidirectional communication through stdin/stdout</li>
5871 | <li>Process-based integration support</li>
5872 | <li>Simple setup and configuration</li>
5873 | <li>Lightweight implementation</li>
5874 | </ul>
5875 | </>
5876 | </Tab>
5877 |
5878 | <Tab title="SSE (WebFlux)">
5879 | <>
5880 | <p>Creates WebFlux-based SSE server transport.<br />Requires the <code>mcp-spring-webflux</code> dependency.</p>
5881 |
5882 | ```java
5883 | @Configuration
5884 | class McpConfig {
5885 | @Bean
5886 | WebFluxSseServerTransport webFluxSseServerTransport(ObjectMapper mapper) {
5887 | return new WebFluxSseServerTransport(mapper, "/mcp/message");
5888 | }
5889 |
5890 | @Bean
5891 | RouterFunction<?> mcpRouterFunction(WebFluxSseServerTransport transport) {
5892 | return transport.getRouterFunction();
5893 | }
5894 | }
5895 | ```
5896 |
5897 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5898 |
5899 | <ul>
5900 | <li>Reactive HTTP streaming with WebFlux</li>
5901 | <li>Concurrent client connections through SSE endpoints</li>
5902 | <li>Message routing and session management</li>
5903 | <li>Graceful shutdown capabilities</li>
5904 | </ul>
5905 | </>
5906 | </Tab>
5907 |
5908 | <Tab title="SSE (WebMvc)">
5909 | <>
5910 | <p>Creates WebMvc-based SSE server transport.<br />Requires the <code>mcp-spring-webmvc</code> dependency.</p>
5911 |
5912 | ```java
5913 | @Configuration
5914 | @EnableWebMvc
5915 | class McpConfig {
5916 | @Bean
5917 | WebMvcSseServerTransport webMvcSseServerTransport(ObjectMapper mapper) {
5918 | return new WebMvcSseServerTransport(mapper, "/mcp/message");
5919 | }
5920 |
5921 | @Bean
5922 | RouterFunction<ServerResponse> mcpRouterFunction(WebMvcSseServerTransport transport) {
5923 | return transport.getRouterFunction();
5924 | }
5925 | }
5926 | ```
5927 |
5928 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5929 |
5930 | <ul>
5931 | <li>Server-side event streaming</li>
5932 | <li>Integration with Spring WebMVC</li>
5933 | <li>Support for traditional web applications</li>
5934 | <li>Synchronous operation handling</li>
5935 | </ul>
5936 | </>
5937 | </Tab>
5938 |
5939 | <Tab title="SSE (Servlet)">
5940 | <>
5941 | <p>
5942 | Creates a Servlet-based SSE server transport. It is included in the core <code>mcp</code> module.<br />
5943 | The <code>HttpServletSseServerTransport</code> can be used with any Servlet container.<br />
5944 | To use it with a Spring Web application, you can register it as a Servlet bean:
5945 | </p>
5946 |
5947 | ```java
5948 | @Configuration
5949 | @EnableWebMvc
5950 | public class McpServerConfig implements WebMvcConfigurer {
5951 |
5952 | @Bean
5953 | public HttpServletSseServerTransport servletSseServerTransport() {
5954 | return new HttpServletSseServerTransport(new ObjectMapper(), "/mcp/message");
5955 | }
5956 |
5957 | @Bean
5958 | public ServletRegistrationBean customServletBean(HttpServletSseServerTransport servlet) {
5959 | return new ServletRegistrationBean(servlet);
5960 | }
5961 | }
5962 | ```
5963 |
5964 | <p>
5965 | Implements the MCP HTTP with SSE transport specification using the traditional Servlet API, providing:
5966 | </p>
5967 |
5968 | <ul>
5969 | <li>Asynchronous message handling using Servlet 6.0 async support</li>
5970 | <li>Session management for multiple client connections</li>
5971 |
5972 | <li>
5973 | Two types of endpoints:
5974 |
5975 | <ul>
5976 | <li>SSE endpoint (<code>/sse</code>) for server-to-client events</li>
5977 | <li>Message endpoint (configurable) for client-to-server requests</li>
5978 | </ul>
5979 | </li>
5980 |
5981 | <li>Error handling and response formatting</li>
5982 | <li>Graceful shutdown support</li>
5983 | </ul>
5984 | </>
5985 | </Tab>
5986 | </Tabs>
5987 |
5988 | ## Server Capabilities
5989 |
5990 | The server can be configured with various capabilities:
5991 |
5992 | ```java
5993 | var capabilities = ServerCapabilities.builder()
5994 | .resources(false, true) // Resource support with list changes notifications
5995 | .tools(true) // Tool support with list changes notifications
5996 | .prompts(true) // Prompt support with list changes notifications
5997 | .logging() // Enable logging support (enabled by default with loging level INFO)
5998 | .build();
5999 | ```
6000 |
6001 | ### Logging Support
6002 |
6003 | The server provides structured logging capabilities that allow sending log messages to clients with different severity levels:
6004 |
6005 | ```java
6006 | // Send a log message to clients
6007 | server.loggingNotification(LoggingMessageNotification.builder()
6008 | .level(LoggingLevel.INFO)
6009 | .logger("custom-logger")
6010 | .data("Custom log message")
6011 | .build());
6012 | ```
6013 |
6014 | Clients can control the minimum logging level they receive through the `mcpClient.setLoggingLevel(level)` request. Messages below the set level will be filtered out.
6015 | Supported logging levels (in order of increasing severity): DEBUG (0), INFO (1), NOTICE (2), WARNING (3), ERROR (4), CRITICAL (5), ALERT (6), EMERGENCY (7)
6016 |
6017 | ### Tool Registration
6018 |
6019 | <Tabs>
6020 | <Tab title="Sync">
6021 | ```java
6022 | // Sync tool registration
6023 | var syncToolRegistration = new McpServerFeatures.SyncToolRegistration(
6024 | new Tool("calculator", "Basic calculator", Map.of(
6025 | "operation", "string",
6026 | "a", "number",
6027 | "b", "number"
6028 | )),
6029 | arguments -> {
6030 | // Tool implementation
6031 | return new CallToolResult(result, false);
6032 | }
6033 | );
6034 | ```
6035 | </Tab>
6036 |
6037 | <Tab title="Async">
6038 | ```java
6039 | // Async tool registration
6040 | var asyncToolRegistration = new McpServerFeatures.AsyncToolRegistration(
6041 | new Tool("calculator", "Basic calculator", Map.of(
6042 | "operation", "string",
6043 | "a", "number",
6044 | "b", "number"
6045 | )),
6046 | arguments -> {
6047 | // Tool implementation
6048 | return Mono.just(new CallToolResult(result, false));
6049 | }
6050 | );
6051 | ```
6052 | </Tab>
6053 | </Tabs>
6054 |
6055 | ### Resource Registration
6056 |
6057 | <Tabs>
6058 | <Tab title="Sync">
6059 | ```java
6060 | // Sync resource registration
6061 | var syncResourceRegistration = new McpServerFeatures.SyncResourceRegistration(
6062 | new Resource("custom://resource", "name", "description", "mime-type", null),
6063 | request -> {
6064 | // Resource read implementation
6065 | return new ReadResourceResult(contents);
6066 | }
6067 | );
6068 | ```
6069 | </Tab>
6070 |
6071 | <Tab title="Async">
6072 | ```java
6073 | // Async resource registration
6074 | var asyncResourceRegistration = new McpServerFeatures.AsyncResourceRegistration(
6075 | new Resource("custom://resource", "name", "description", "mime-type", null),
6076 | request -> {
6077 | // Resource read implementation
6078 | return Mono.just(new ReadResourceResult(contents));
6079 | }
6080 | );
6081 | ```
6082 | </Tab>
6083 | </Tabs>
6084 |
6085 | ### Prompt Registration
6086 |
6087 | <Tabs>
6088 | <Tab title="Sync">
6089 | ```java
6090 | // Sync prompt registration
6091 | var syncPromptRegistration = new McpServerFeatures.SyncPromptRegistration(
6092 | new Prompt("greeting", "description", List.of(
6093 | new PromptArgument("name", "description", true)
6094 | )),
6095 | request -> {
6096 | // Prompt implementation
6097 | return new GetPromptResult(description, messages);
6098 | }
6099 | );
6100 | ```
6101 | </Tab>
6102 |
6103 | <Tab title="Async">
6104 | ```java
6105 | // Async prompt registration
6106 | var asyncPromptRegistration = new McpServerFeatures.AsyncPromptRegistration(
6107 | new Prompt("greeting", "description", List.of(
6108 | new PromptArgument("name", "description", true)
6109 | )),
6110 | request -> {
6111 | // Prompt implementation
6112 | return Mono.just(new GetPromptResult(description, messages));
6113 | }
6114 | );
6115 | ```
6116 | </Tab>
6117 | </Tabs>
6118 |
6119 | ## Error Handling
6120 |
6121 | 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.
6122 |
6123 |
6124 | # Building MCP with LLMs
6125 | Source: https://modelcontextprotocol.io/tutorials/building-mcp-with-llms
6126 |
6127 | Speed up your MCP development using LLMs such as Claude!
6128 |
6129 | 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.
6130 |
6131 | ## Preparing the documentation
6132 |
6133 | Before starting, gather the necessary documentation to help Claude understand MCP:
6134 |
6135 | 1. Visit [https://modelcontextprotocol.io/llms-full.txt](https://modelcontextprotocol.io/llms-full.txt) and copy the full documentation text
6136 | 2. Navigate to either the [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) or [Python SDK repository](https://github.com/modelcontextprotocol/python-sdk)
6137 | 3. Copy the README files and other relevant documentation
6138 | 4. Paste these documents into your conversation with Claude
6139 |
6140 | ## Describing your server
6141 |
6142 | Once you've provided the documentation, clearly describe to Claude what kind of server you want to build. Be specific about:
6143 |
6144 | * What resources your server will expose
6145 | * What tools it will provide
6146 | * Any prompts it should offer
6147 | * What external systems it needs to interact with
6148 |
6149 | For example:
6150 |
6151 | ```
6152 | Build an MCP server that:
6153 | - Connects to my company's PostgreSQL database
6154 | - Exposes table schemas as resources
6155 | - Provides tools for running read-only SQL queries
6156 | - Includes prompts for common data analysis tasks
6157 | ```
6158 |
6159 | ## Working with Claude
6160 |
6161 | When working with Claude on MCP servers:
6162 |
6163 | 1. Start with the core functionality first, then iterate to add more features
6164 | 2. Ask Claude to explain any parts of the code you don't understand
6165 | 3. Request modifications or improvements as needed
6166 | 4. Have Claude help you test the server and handle edge cases
6167 |
6168 | Claude can help implement all the key MCP features:
6169 |
6170 | * Resource management and exposure
6171 | * Tool definitions and implementations
6172 | * Prompt templates and handlers
6173 | * Error handling and logging
6174 | * Connection and transport setup
6175 |
6176 | ## Best practices
6177 |
6178 | When building MCP servers with Claude:
6179 |
6180 | * Break down complex servers into smaller pieces
6181 | * Test each component thoroughly before moving on
6182 | * Keep security in mind - validate inputs and limit access appropriately
6183 | * Document your code well for future maintenance
6184 | * Follow MCP protocol specifications carefully
6185 |
6186 | ## Next steps
6187 |
6188 | After Claude helps you build your server:
6189 |
6190 | 1. Review the generated code carefully
6191 | 2. Test the server with the MCP Inspector tool
6192 | 3. Connect it to Claude.app or other MCP clients
6193 | 4. Iterate based on real usage and feedback
6194 |
6195 | Remember that Claude can help you modify and improve your server as requirements change over time.
6196 |
6197 | Need more guidance? Just ask Claude specific questions about implementing MCP features or troubleshooting issues that arise.
6198 |
6199 |
```