This is page 2 of 2. Use http://codebase.md/illuminaresolutions/n8n-mcp-server?lines=true&page={x} to view the full context.
# Directory Structure
```
├── .gitignore
├── DEVELOPMENT.md
├── LICENSE
├── LLM_GUIDE.md
├── N8N_API.yml
├── package-lock.json
├── package.json
├── README.md
├── smithery.yml
├── src
│ └── index.ts
├── tsconfig.json
└── TYPEKIT_MCP_SDK.md
```
# Files
--------------------------------------------------------------------------------
/LLM_GUIDE.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 | | [Bee Agent Framework][Bee Agent 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 |
31 | [Claude]: https://claude.ai/download
32 |
33 | [Cursor]: https://cursor.com
34 |
35 | [Zed]: https://zed.dev
36 |
37 | [Cody]: https://sourcegraph.com/cody
38 |
39 | [Genkit]: https://github.com/firebase/genkit
40 |
41 | [Continue]: https://github.com/continuedev/continue
42 |
43 | [GenAIScript]: https://microsoft.github.io/genaiscript/reference/scripts/mcp-tools/
44 |
45 | [Cline]: https://github.com/cline/cline
46 |
47 | [LibreChat]: https://github.com/danny-avila/LibreChat
48 |
49 | [TheiaAI/TheiaIDE]: https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/
50 |
51 | [Superinterface]: https://superinterface.ai
52 |
53 | [5ire]: https://github.com/nanbingxyz/5ire
54 |
55 | [Bee Agent Framework]: https://i-am-bee.github.io/bee-agent-framework
56 |
57 | [mcp-agent]: https://github.com/lastmile-ai/mcp-agent
58 |
59 | [Mcp.el]: https://github.com/lizqwerscott/mcp.el
60 |
61 | [Roo Code]: https://roocode.com
62 |
63 | [Goose]: https://block.github.io/goose/docs/goose-architecture/#interoperability-with-extensions
64 |
65 | [Windsurf]: https://codeium.com/windsurf
66 |
67 | [Resources]: https://modelcontextprotocol.io/docs/concepts/resources
68 |
69 | [Prompts]: https://modelcontextprotocol.io/docs/concepts/prompts
70 |
71 | [Tools]: https://modelcontextprotocol.io/docs/concepts/tools
72 |
73 | [Sampling]: https://modelcontextprotocol.io/docs/concepts/sampling
74 |
75 | ## Client details
76 |
77 | ### Claude Desktop App
78 |
79 | The Claude desktop application provides comprehensive support for MCP, enabling deep integration with local tools and data sources.
80 |
81 | **Key features:**
82 |
83 | * Full support for resources, allowing attachment of local files and data
84 | * Support for prompt templates
85 | * Tool integration for executing commands and scripts
86 | * Local server connections for enhanced privacy and security
87 |
88 | > ⓘ Note: The Claude.ai web application does not currently support MCP. MCP features are only available in the desktop application.
89 |
90 | ### 5ire
91 |
92 | [5ire](https://github.com/nanbingxyz/5ire) is an open source cross-platform desktop AI assistant that supports tools through MCP servers.
93 |
94 | **Key features:**
95 |
96 | * Built-in MCP servers can be quickly enabled and disabled.
97 | * Users can add more servers by modifying the configuration file.
98 | * It is open-source and user-friendly, suitable for beginners.
99 | * Future support for MCP will be continuously improved.
100 |
101 | ### Bee Agent Framework
102 |
103 | [Bee Agent Framework](https://i-am-bee.github.io/bee-agent-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.
104 |
105 | **Key features:**
106 |
107 | * Seamlessly incorporate MCP tools into agentic workflows.
108 | * Quickly instantiate framework-native tools from connected MCP client(s).
109 | * Planned future support for agentic MCP capabilities.
110 |
111 | **Learn more:**
112 |
113 | * [Example of using MCP tools in agentic workflow](https://i-am-bee.github.io/bee-agent-framework/#/tools?id=using-the-mcptool-class)
114 |
115 | ### Cline
116 |
117 | [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.
118 |
119 | **Key features:**
120 |
121 | * Create and add tools through natural language (e.g. "add a tool that searches the web")
122 | * Share custom MCP servers Cline creates with others via the `~/Documents/Cline/MCP` directory
123 | * Displays configured MCP servers along with their tools, resources, and any error logs
124 |
125 | ### Continue
126 |
127 | [Continue](https://github.com/continuedev/continue) is an open-source AI code assistant, with built-in support for all MCP features.
128 |
129 | **Key features**
130 |
131 | * Type "@" to mention MCP resources
132 | * Prompt templates surface as slash commands
133 | * Use both built-in and MCP tools directly in chat
134 | * Supports VS Code and JetBrains IDEs, with any LLM
135 |
136 | ### Cursor
137 |
138 | [Cursor](https://docs.cursor.com/advanced/model-context-protocol) is an AI code editor.
139 |
140 | **Key Features**:
141 |
142 | * Support for MCP tools in Cursor Composer
143 | * Support for both STDIO and SSE
144 |
145 | ### Emacs Mcp
146 |
147 | [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.
148 |
149 | **Key features:**
150 |
151 | * Provides MCP tool support for Emacs.
152 |
153 | ### Firebase Genkit
154 |
155 | [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.
156 |
157 | **Key features:**
158 |
159 | * Client support for tools and prompts (resources partially supported)
160 | * Rich discovery with support in Genkit's Dev UI playground
161 | * Seamless interoperability with Genkit's existing tools and prompts
162 | * Works across a wide variety of GenAI models from top providers
163 |
164 | ### GenAIScript
165 |
166 | Programmatically assemble prompts for LLMs using [GenAIScript](https://microsoft.github.io/genaiscript/) (in JavaScript). Orchestrate LLMs, tools, and data in JavaScript.
167 |
168 | **Key features:**
169 |
170 | * JavaScript toolbox to work with prompts
171 | * Abstraction to make it easy and productive
172 | * Seamless Visual Studio Code integration
173 |
174 | ### Goose
175 |
176 | [Goose](https://github.com/block/goose) is an open source AI agent that supercharges your software development by automating coding tasks.
177 |
178 | **Key features:**
179 |
180 | * Expose MCP functionality to Goose through tools.
181 | * MCPs can be installed directly via the [extensions directory](https://block.github.io/goose/v1/extensions/), CLI, or UI.
182 | * Goose allows you to extend its functionality by [building your own MCP servers](https://block.github.io/goose/docs/tutorials/custom-extensions).
183 | * Includes built-in tools for development, web scraping, automation, memory, and integrations with JetBrains and Google Drive.
184 |
185 | ### LibreChat
186 |
187 | [LibreChat](https://github.com/danny-avila/LibreChat) is an open-source, customizable AI chat UI that supports multiple AI providers, now including MCP integration.
188 |
189 | **Key features:**
190 |
191 | * Extend current tool ecosystem, including [Code Interpreter](https://www.librechat.ai/docs/features/code_interpreter) and Image generation tools, through MCP servers
192 | * Add tools to customizable [Agents](https://www.librechat.ai/docs/features/agents), using a variety of LLMs from top providers
193 | * Open-source and self-hostable, with secure multi-user support
194 | * Future roadmap includes expanded MCP feature support
195 |
196 | ### mcp-agent
197 |
198 | [mcp-agent] is a simple, composable framework to build agents using Model Context Protocol.
199 |
200 | **Key features:**
201 |
202 | * Automatic connection management of MCP servers.
203 | * Expose tools from multiple servers to an LLM.
204 | * Implements every pattern defined in [Building Effective Agents](https://www.anthropic.com/research/building-effective-agents).
205 | * Supports workflow pause/resume signals, such as waiting for human feedback.
206 |
207 | ### Roo Code
208 |
209 | [Roo Code](https://roocode.com) enables AI coding assistance via MCP.
210 |
211 | **Key features:**
212 |
213 | * Support for MCP tools and resources
214 | * Integration with development workflows
215 | * Extensible AI capabilities
216 |
217 | ### Sourcegraph Cody
218 |
219 | [Cody](https://openctx.org/docs/providers/modelcontextprotocol) is Sourcegraph's AI coding assistant, which implements MCP through OpenCTX.
220 |
221 | **Key features:**
222 |
223 | * Support for MCP resources
224 | * Integration with Sourcegraph's code intelligence
225 | * Uses OpenCTX as an abstraction layer
226 | * Future support planned for additional MCP features
227 |
228 | ### Superinterface
229 |
230 | [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.
231 |
232 | **Key features:**
233 |
234 | * Use tools from MCP servers in assistants embedded via React components or script tags
235 | * SSE transport support
236 | * Use any AI model from any AI provider (OpenAI, Anthropic, Ollama, others)
237 |
238 | ### TheiaAI/TheiaIDE
239 |
240 | [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.
241 |
242 | **Key features:**
243 |
244 | * **Tool Integration**: Theia AI enables AI agents, including those in the Theia IDE, to utilize MCP servers for seamless tool interaction.
245 | * **Customizable Prompts**: The Theia IDE allows users to define and adapt prompts, dynamically integrating MCP servers for tailored workflows.
246 | * **Custom agents**: The Theia IDE supports creating custom agents that leverage MCP capabilities, enabling users to design dedicated workflows on the fly.
247 |
248 | Theia AI and Theia IDE's MCP integration provide users with flexibility, making them powerful platforms for exploring and adapting MCP.
249 |
250 | **Learn more:**
251 |
252 | * [Theia IDE and Theia AI MCP Announcement](https://eclipsesource.com/blogs/2024/12/19/theia-ide-and-theia-ai-support-mcp/)
253 | * [Download the AI-powered Theia IDE](https://theia-ide.org/)
254 |
255 | ### Windsurf Editor
256 |
257 | [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.
258 |
259 | **Key features:**
260 |
261 | * Revolutionary AI Flow paradigm for human-AI collaboration
262 | * Intelligent code generation and understanding
263 | * Rich development tools with multi-model support
264 |
265 | ### Zed
266 |
267 | [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.
268 |
269 | **Key features:**
270 |
271 | * Prompt templates surface as slash commands in the editor
272 | * Tool integration for enhanced coding workflows
273 | * Tight integration with editor features and workspace context
274 | * Does not support MCP resources
275 |
276 | ## Adding MCP support to your application
277 |
278 | 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.
279 |
280 | Benefits of adding MCP support:
281 |
282 | * Enable users to bring their own context and tools
283 | * Join a growing ecosystem of interoperable AI applications
284 | * Provide users with flexible integration options
285 | * Support local-first AI workflows
286 |
287 | 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)
288 |
289 | ## Updates and corrections
290 |
291 | 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).
292 |
293 |
294 | # Contributing
295 | Source: https://modelcontextprotocol.io/development/contributing
296 |
297 | How to participate in Model Context Protocol development
298 |
299 | 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.
300 |
301 | All contributors must adhere to our [Code of Conduct](https://github.com/modelcontextprotocol/.github/blob/main/CODE_OF_CONDUCT.md).
302 |
303 | For questions and discussions, please use [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions).
304 |
305 |
306 | # Roadmap
307 | Source: https://modelcontextprotocol.io/development/roadmap
308 |
309 | Our plans for evolving Model Context Protocol (H1 2025)
310 |
311 | 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.
312 |
313 | <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>
314 |
315 | We encourage community participation! Each section links to relevant discussions where you can learn more and contribute your thoughts.
316 |
317 | ## Remote MCP Support
318 |
319 | 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:
320 |
321 | * [**Authentication & Authorization**](https://github.com/modelcontextprotocol/specification/discussions/64): Adding standardized auth capabilities, particularly focused on OAuth 2.0 support.
322 |
323 | * [**Service Discovery**](https://github.com/modelcontextprotocol/specification/discussions/69): Defining how clients can discover and connect to remote MCP servers.
324 |
325 | * [**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.
326 |
327 | ## Reference Implementations
328 |
329 | To help developers build with MCP, we want to offer documentation for:
330 |
331 | * **Client Examples**: Comprehensive reference client implementation(s), demonstrating all protocol features
332 | * **Protocol Drafting**: Streamlined process for proposing and incorporating new protocol features
333 |
334 | ## Distribution & Discovery
335 |
336 | Looking ahead, we're exploring ways to make MCP servers more accessible. Some areas we may investigate include:
337 |
338 | * **Package Management**: Standardized packaging format for MCP servers
339 | * **Installation Tools**: Simplified server installation across MCP clients
340 | * **Sandboxing**: Improved security through server isolation
341 | * **Server Registry**: A common directory for discovering available MCP servers
342 |
343 | ## Agent Support
344 |
345 | We're expanding MCP's capabilities for [complex agentic workflows](https://github.com/modelcontextprotocol/specification/discussions/111), particularly focusing on:
346 |
347 | * [**Hierarchical Agent Systems**](https://github.com/modelcontextprotocol/specification/discussions/94): Improved support for trees of agents through namespacing and topology awareness.
348 |
349 | * [**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.
350 |
351 | * [**Streaming Results**](https://github.com/modelcontextprotocol/specification/issues/117): Real-time updates from long-running agent operations.
352 |
353 | ## Broader Ecosystem
354 |
355 | We're also invested in:
356 |
357 | * **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.
358 | * [**Additional Modalities**](https://github.com/modelcontextprotocol/specification/discussions/88): Expanding beyond text to support audio, video, and other formats.
359 | * \[**Standardization**] Considering standardization through a standardization body.
360 |
361 | ## Get Involved
362 |
363 | 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.
364 |
365 |
366 | # What's New
367 | Source: https://modelcontextprotocol.io/development/updates
368 |
369 | The latest updates and improvements to MCP
370 |
371 | <Update label="2025-02-14" description="Java SDK released">
372 | * We're excited to announce that the Java SDK developed by Spring AI at VMware Tanzu is now
373 | the official [Java SDK](https://github.com/modelcontextprotocol/java-sdk) for MCP.
374 | This joins our existing Kotlin SDK in our growing list of supported languages.
375 | The Spring AI team will maintain the SDK as an integral part of the Model Context Protocol
376 | organization. We're thrilled to welcome them to the MCP community!
377 | </Update>
378 |
379 | <Update label="2025-01-27" description="Python SDK 1.2.1">
380 | * Version [1.2.1](https://github.com/modelcontextprotocol/python-sdk/releases/tag/v1.2.1) of the MCP Python SDK has been released,
381 | delivering important stability improvements and bug fixes.
382 | </Update>
383 |
384 | <Update label="2025-01-18" description="SDK and Server Improvements">
385 | * Simplified, express-like API in the [TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk)
386 | * Added 8 new clients to the [clients page](https://modelcontextprotocol.io/clients)
387 | </Update>
388 |
389 | <Update label="2025-01-03" description="SDK and Server Improvements">
390 | * FastMCP API in the [Python SDK](https://github.com/modelcontextprotocol/python-sdk)
391 | * Dockerized MCP servers in the [servers repo](https://github.com/modelcontextprotocol/servers)
392 | </Update>
393 |
394 | <Update label="2024-12-21" description="Kotlin SDK released">
395 | * Jetbrains released a Kotlin SDK for MCP!
396 | * For a sample MCP Kotlin server, check out [this repository](https://github.com/modelcontextprotocol/kotlin-sdk/tree/main/samples/kotlin-mcp-server)
397 | </Update>
398 |
399 |
400 | # Core architecture
401 | Source: https://modelcontextprotocol.io/docs/concepts/architecture
402 |
403 | Understand how MCP connects clients, servers, and LLMs
404 |
405 | 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.
406 |
407 | ## Overview
408 |
409 | MCP follows a client-server architecture where:
410 |
411 | * **Hosts** are LLM applications (like Claude Desktop or IDEs) that initiate connections
412 | * **Clients** maintain 1:1 connections with servers, inside the host application
413 | * **Servers** provide context, tools, and prompts to clients
414 |
415 | ```mermaid
416 | flowchart LR
417 | subgraph " Host (e.g., Claude Desktop) "
418 | client1[MCP Client]
419 | client2[MCP Client]
420 | end
421 | subgraph "Server Process"
422 | server1[MCP Server]
423 | end
424 | subgraph "Server Process"
425 | server2[MCP Server]
426 | end
427 |
428 | client1 <-->|Transport Layer| server1
429 | client2 <-->|Transport Layer| server2
430 | ```
431 |
432 | ## Core components
433 |
434 | ### Protocol layer
435 |
436 | The protocol layer handles message framing, request/response linking, and high-level communication patterns.
437 |
438 | <Tabs>
439 | <Tab title="TypeScript">
440 | ```typescript
441 | class Protocol<Request, Notification, Result> {
442 | // Handle incoming requests
443 | setRequestHandler<T>(schema: T, handler: (request: T, extra: RequestHandlerExtra) => Promise<Result>): void
444 |
445 | // Handle incoming notifications
446 | setNotificationHandler<T>(schema: T, handler: (notification: T) => Promise<void>): void
447 |
448 | // Send requests and await responses
449 | request<T>(request: Request, schema: T, options?: RequestOptions): Promise<T>
450 |
451 | // Send one-way notifications
452 | notification(notification: Notification): Promise<void>
453 | }
454 | ```
455 | </Tab>
456 |
457 | <Tab title="Python">
458 | ```python
459 | class Session(BaseSession[RequestT, NotificationT, ResultT]):
460 | async def send_request(
461 | self,
462 | request: RequestT,
463 | result_type: type[Result]
464 | ) -> Result:
465 | """
466 | Send request and wait for response. Raises McpError if response contains error.
467 | """
468 | # Request handling implementation
469 |
470 | async def send_notification(
471 | self,
472 | notification: NotificationT
473 | ) -> None:
474 | """Send one-way notification that doesn't expect response."""
475 | # Notification handling implementation
476 |
477 | async def _received_request(
478 | self,
479 | responder: RequestResponder[ReceiveRequestT, ResultT]
480 | ) -> None:
481 | """Handle incoming request from other side."""
482 | # Request handling implementation
483 |
484 | async def _received_notification(
485 | self,
486 | notification: ReceiveNotificationT
487 | ) -> None:
488 | """Handle incoming notification from other side."""
489 | # Notification handling implementation
490 | ```
491 | </Tab>
492 | </Tabs>
493 |
494 | Key classes include:
495 |
496 | * `Protocol`
497 | * `Client`
498 | * `Server`
499 |
500 | ### Transport layer
501 |
502 | The transport layer handles the actual communication between clients and servers. MCP supports multiple transport mechanisms:
503 |
504 | 1. **Stdio transport**
505 | * Uses standard input/output for communication
506 | * Ideal for local processes
507 |
508 | 2. **HTTP with SSE transport**
509 | * Uses Server-Sent Events for server-to-client messages
510 | * HTTP POST for client-to-server messages
511 |
512 | 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.
513 |
514 | ### Message types
515 |
516 | MCP has these main types of messages:
517 |
518 | 1. **Requests** expect a response from the other side:
519 | ```typescript
520 | interface Request {
521 | method: string;
522 | params?: { ... };
523 | }
524 | ```
525 |
526 | 2. **Results** are successful responses to requests:
527 | ```typescript
528 | interface Result {
529 | [key: string]: unknown;
530 | }
531 | ```
532 |
533 | 3. **Errors** indicate that a request failed:
534 | ```typescript
535 | interface Error {
536 | code: number;
537 | message: string;
538 | data?: unknown;
539 | }
540 | ```
541 |
542 | 4. **Notifications** are one-way messages that don't expect a response:
543 | ```typescript
544 | interface Notification {
545 | method: string;
546 | params?: { ... };
547 | }
548 | ```
549 |
550 | ## Connection lifecycle
551 |
552 | ### 1. Initialization
553 |
554 | ```mermaid
555 | sequenceDiagram
556 | participant Client
557 | participant Server
558 |
559 | Client->>Server: initialize request
560 | Server->>Client: initialize response
561 | Client->>Server: initialized notification
562 |
563 | Note over Client,Server: Connection ready for use
564 | ```
565 |
566 | 1. Client sends `initialize` request with protocol version and capabilities
567 | 2. Server responds with its protocol version and capabilities
568 | 3. Client sends `initialized` notification as acknowledgment
569 | 4. Normal message exchange begins
570 |
571 | ### 2. Message exchange
572 |
573 | After initialization, the following patterns are supported:
574 |
575 | * **Request-Response**: Client or server sends requests, the other responds
576 | * **Notifications**: Either party sends one-way messages
577 |
578 | ### 3. Termination
579 |
580 | Either party can terminate the connection:
581 |
582 | * Clean shutdown via `close()`
583 | * Transport disconnection
584 | * Error conditions
585 |
586 | ## Error handling
587 |
588 | MCP defines these standard error codes:
589 |
590 | ```typescript
591 | enum ErrorCode {
592 | // Standard JSON-RPC error codes
593 | ParseError = -32700,
594 | InvalidRequest = -32600,
595 | MethodNotFound = -32601,
596 | InvalidParams = -32602,
597 | InternalError = -32603
598 | }
599 | ```
600 |
601 | SDKs and applications can define their own error codes above -32000.
602 |
603 | Errors are propagated through:
604 |
605 | * Error responses to requests
606 | * Error events on transports
607 | * Protocol-level error handlers
608 |
609 | ## Implementation example
610 |
611 | Here's a basic example of implementing an MCP server:
612 |
613 | <Tabs>
614 | <Tab title="TypeScript">
615 | ```typescript
616 | import { Server } from "@modelcontextprotocol/sdk/server/index.js";
617 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
618 |
619 | const server = new Server({
620 | name: "example-server",
621 | version: "1.0.0"
622 | }, {
623 | capabilities: {
624 | resources: {}
625 | }
626 | });
627 |
628 | // Handle requests
629 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
630 | return {
631 | resources: [
632 | {
633 | uri: "example://resource",
634 | name: "Example Resource"
635 | }
636 | ]
637 | };
638 | });
639 |
640 | // Connect transport
641 | const transport = new StdioServerTransport();
642 | await server.connect(transport);
643 | ```
644 | </Tab>
645 |
646 | <Tab title="Python">
647 | ```python
648 | import asyncio
649 | import mcp.types as types
650 | from mcp.server import Server
651 | from mcp.server.stdio import stdio_server
652 |
653 | app = Server("example-server")
654 |
655 | @app.list_resources()
656 | async def list_resources() -> list[types.Resource]:
657 | return [
658 | types.Resource(
659 | uri="example://resource",
660 | name="Example Resource"
661 | )
662 | ]
663 |
664 | async def main():
665 | async with stdio_server() as streams:
666 | await app.run(
667 | streams[0],
668 | streams[1],
669 | app.create_initialization_options()
670 | )
671 |
672 | if __name__ == "__main__":
673 | asyncio.run(main)
674 | ```
675 | </Tab>
676 | </Tabs>
677 |
678 | ## Best practices
679 |
680 | ### Transport selection
681 |
682 | 1. **Local communication**
683 | * Use stdio transport for local processes
684 | * Efficient for same-machine communication
685 | * Simple process management
686 |
687 | 2. **Remote communication**
688 | * Use SSE for scenarios requiring HTTP compatibility
689 | * Consider security implications including authentication and authorization
690 |
691 | ### Message handling
692 |
693 | 1. **Request processing**
694 | * Validate inputs thoroughly
695 | * Use type-safe schemas
696 | * Handle errors gracefully
697 | * Implement timeouts
698 |
699 | 2. **Progress reporting**
700 | * Use progress tokens for long operations
701 | * Report progress incrementally
702 | * Include total progress when known
703 |
704 | 3. **Error management**
705 | * Use appropriate error codes
706 | * Include helpful error messages
707 | * Clean up resources on errors
708 |
709 | ## Security considerations
710 |
711 | 1. **Transport security**
712 | * Use TLS for remote connections
713 | * Validate connection origins
714 | * Implement authentication when needed
715 |
716 | 2. **Message validation**
717 | * Validate all incoming messages
718 | * Sanitize inputs
719 | * Check message size limits
720 | * Verify JSON-RPC format
721 |
722 | 3. **Resource protection**
723 | * Implement access controls
724 | * Validate resource paths
725 | * Monitor resource usage
726 | * Rate limit requests
727 |
728 | 4. **Error handling**
729 | * Don't leak sensitive information
730 | * Log security-relevant errors
731 | * Implement proper cleanup
732 | * Handle DoS scenarios
733 |
734 | ## Debugging and monitoring
735 |
736 | 1. **Logging**
737 | * Log protocol events
738 | * Track message flow
739 | * Monitor performance
740 | * Record errors
741 |
742 | 2. **Diagnostics**
743 | * Implement health checks
744 | * Monitor connection state
745 | * Track resource usage
746 | * Profile performance
747 |
748 | 3. **Testing**
749 | * Test different transports
750 | * Verify error handling
751 | * Check edge cases
752 | * Load test servers
753 |
754 |
755 | # Prompts
756 | Source: https://modelcontextprotocol.io/docs/concepts/prompts
757 |
758 | Create reusable prompt templates and workflows
759 |
760 | 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.
761 |
762 | <Note>
763 | 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.
764 | </Note>
765 |
766 | ## Overview
767 |
768 | Prompts in MCP are predefined templates that can:
769 |
770 | * Accept dynamic arguments
771 | * Include context from resources
772 | * Chain multiple interactions
773 | * Guide specific workflows
774 | * Surface as UI elements (like slash commands)
775 |
776 | ## Prompt structure
777 |
778 | Each prompt is defined with:
779 |
780 | ```typescript
781 | {
782 | name: string; // Unique identifier for the prompt
783 | description?: string; // Human-readable description
784 | arguments?: [ // Optional list of arguments
785 | {
786 | name: string; // Argument identifier
787 | description?: string; // Argument description
788 | required?: boolean; // Whether argument is required
789 | }
790 | ]
791 | }
792 | ```
793 |
794 | ## Discovering prompts
795 |
796 | Clients can discover available prompts through the `prompts/list` endpoint:
797 |
798 | ```typescript
799 | // Request
800 | {
801 | method: "prompts/list"
802 | }
803 |
804 | // Response
805 | {
806 | prompts: [
807 | {
808 | name: "analyze-code",
809 | description: "Analyze code for potential improvements",
810 | arguments: [
811 | {
812 | name: "language",
813 | description: "Programming language",
814 | required: true
815 | }
816 | ]
817 | }
818 | ]
819 | }
820 | ```
821 |
822 | ## Using prompts
823 |
824 | To use a prompt, clients make a `prompts/get` request:
825 |
826 | ````typescript
827 | // Request
828 | {
829 | method: "prompts/get",
830 | params: {
831 | name: "analyze-code",
832 | arguments: {
833 | language: "python"
834 | }
835 | }
836 | }
837 |
838 | // Response
839 | {
840 | description: "Analyze Python code for potential improvements",
841 | messages: [
842 | {
843 | role: "user",
844 | content: {
845 | type: "text",
846 | 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```"
847 | }
848 | }
849 | ]
850 | }
851 | ````
852 |
853 | ## Dynamic prompts
854 |
855 | Prompts can be dynamic and include:
856 |
857 | ### Embedded resource context
858 |
859 | ```json
860 | {
861 | "name": "analyze-project",
862 | "description": "Analyze project logs and code",
863 | "arguments": [
864 | {
865 | "name": "timeframe",
866 | "description": "Time period to analyze logs",
867 | "required": true
868 | },
869 | {
870 | "name": "fileUri",
871 | "description": "URI of code file to review",
872 | "required": true
873 | }
874 | ]
875 | }
876 | ```
877 |
878 | When handling the `prompts/get` request:
879 |
880 | ```json
881 | {
882 | "messages": [
883 | {
884 | "role": "user",
885 | "content": {
886 | "type": "text",
887 | "text": "Analyze these system logs and the code file for any issues:"
888 | }
889 | },
890 | {
891 | "role": "user",
892 | "content": {
893 | "type": "resource",
894 | "resource": {
895 | "uri": "logs://recent?timeframe=1h",
896 | "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",
897 | "mimeType": "text/plain"
898 | }
899 | }
900 | },
901 | {
902 | "role": "user",
903 | "content": {
904 | "type": "resource",
905 | "resource": {
906 | "uri": "file:///path/to/code.py",
907 | "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",
908 | "mimeType": "text/x-python"
909 | }
910 | }
911 | }
912 | ]
913 | }
914 | ```
915 |
916 | ### Multi-step workflows
917 |
918 | ```typescript
919 | const debugWorkflow = {
920 | name: "debug-error",
921 | async getMessages(error: string) {
922 | return [
923 | {
924 | role: "user",
925 | content: {
926 | type: "text",
927 | text: `Here's an error I'm seeing: ${error}`
928 | }
929 | },
930 | {
931 | role: "assistant",
932 | content: {
933 | type: "text",
934 | text: "I'll help analyze this error. What have you tried so far?"
935 | }
936 | },
937 | {
938 | role: "user",
939 | content: {
940 | type: "text",
941 | text: "I've tried restarting the service, but the error persists."
942 | }
943 | }
944 | ];
945 | }
946 | };
947 | ```
948 |
949 | ## Example implementation
950 |
951 | Here's a complete example of implementing prompts in an MCP server:
952 |
953 | <Tabs>
954 | <Tab title="TypeScript">
955 | ```typescript
956 | import { Server } from "@modelcontextprotocol/sdk/server";
957 | import {
958 | ListPromptsRequestSchema,
959 | GetPromptRequestSchema
960 | } from "@modelcontextprotocol/sdk/types";
961 |
962 | const PROMPTS = {
963 | "git-commit": {
964 | name: "git-commit",
965 | description: "Generate a Git commit message",
966 | arguments: [
967 | {
968 | name: "changes",
969 | description: "Git diff or description of changes",
970 | required: true
971 | }
972 | ]
973 | },
974 | "explain-code": {
975 | name: "explain-code",
976 | description: "Explain how code works",
977 | arguments: [
978 | {
979 | name: "code",
980 | description: "Code to explain",
981 | required: true
982 | },
983 | {
984 | name: "language",
985 | description: "Programming language",
986 | required: false
987 | }
988 | ]
989 | }
990 | };
991 |
992 | const server = new Server({
993 | name: "example-prompts-server",
994 | version: "1.0.0"
995 | }, {
996 | capabilities: {
997 | prompts: {}
998 | }
999 | });
1000 |
1001 | // List available prompts
1002 | server.setRequestHandler(ListPromptsRequestSchema, async () => {
1003 | return {
1004 | prompts: Object.values(PROMPTS)
1005 | };
1006 | });
1007 |
1008 | // Get specific prompt
1009 | server.setRequestHandler(GetPromptRequestSchema, async (request) => {
1010 | const prompt = PROMPTS[request.params.name];
1011 | if (!prompt) {
1012 | throw new Error(`Prompt not found: ${request.params.name}`);
1013 | }
1014 |
1015 | if (request.params.name === "git-commit") {
1016 | return {
1017 | messages: [
1018 | {
1019 | role: "user",
1020 | content: {
1021 | type: "text",
1022 | text: `Generate a concise but descriptive commit message for these changes:\n\n${request.params.arguments?.changes}`
1023 | }
1024 | }
1025 | ]
1026 | };
1027 | }
1028 |
1029 | if (request.params.name === "explain-code") {
1030 | const language = request.params.arguments?.language || "Unknown";
1031 | return {
1032 | messages: [
1033 | {
1034 | role: "user",
1035 | content: {
1036 | type: "text",
1037 | text: `Explain how this ${language} code works:\n\n${request.params.arguments?.code}`
1038 | }
1039 | }
1040 | ]
1041 | };
1042 | }
1043 |
1044 | throw new Error("Prompt implementation not found");
1045 | });
1046 | ```
1047 | </Tab>
1048 |
1049 | <Tab title="Python">
1050 | ```python
1051 | from mcp.server import Server
1052 | import mcp.types as types
1053 |
1054 | # Define available prompts
1055 | PROMPTS = {
1056 | "git-commit": types.Prompt(
1057 | name="git-commit",
1058 | description="Generate a Git commit message",
1059 | arguments=[
1060 | types.PromptArgument(
1061 | name="changes",
1062 | description="Git diff or description of changes",
1063 | required=True
1064 | )
1065 | ],
1066 | ),
1067 | "explain-code": types.Prompt(
1068 | name="explain-code",
1069 | description="Explain how code works",
1070 | arguments=[
1071 | types.PromptArgument(
1072 | name="code",
1073 | description="Code to explain",
1074 | required=True
1075 | ),
1076 | types.PromptArgument(
1077 | name="language",
1078 | description="Programming language",
1079 | required=False
1080 | )
1081 | ],
1082 | )
1083 | }
1084 |
1085 | # Initialize server
1086 | app = Server("example-prompts-server")
1087 |
1088 | @app.list_prompts()
1089 | async def list_prompts() -> list[types.Prompt]:
1090 | return list(PROMPTS.values())
1091 |
1092 | @app.get_prompt()
1093 | async def get_prompt(
1094 | name: str, arguments: dict[str, str] | None = None
1095 | ) -> types.GetPromptResult:
1096 | if name not in PROMPTS:
1097 | raise ValueError(f"Prompt not found: {name}")
1098 |
1099 | if name == "git-commit":
1100 | changes = arguments.get("changes") if arguments else ""
1101 | return types.GetPromptResult(
1102 | messages=[
1103 | types.PromptMessage(
1104 | role="user",
1105 | content=types.TextContent(
1106 | type="text",
1107 | text=f"Generate a concise but descriptive commit message "
1108 | f"for these changes:\n\n{changes}"
1109 | )
1110 | )
1111 | ]
1112 | )
1113 |
1114 | if name == "explain-code":
1115 | code = arguments.get("code") if arguments else ""
1116 | language = arguments.get("language", "Unknown") if arguments else "Unknown"
1117 | return types.GetPromptResult(
1118 | messages=[
1119 | types.PromptMessage(
1120 | role="user",
1121 | content=types.TextContent(
1122 | type="text",
1123 | text=f"Explain how this {language} code works:\n\n{code}"
1124 | )
1125 | )
1126 | ]
1127 | )
1128 |
1129 | raise ValueError("Prompt implementation not found")
1130 | ```
1131 | </Tab>
1132 | </Tabs>
1133 |
1134 | ## Best practices
1135 |
1136 | When implementing prompts:
1137 |
1138 | 1. Use clear, descriptive prompt names
1139 | 2. Provide detailed descriptions for prompts and arguments
1140 | 3. Validate all required arguments
1141 | 4. Handle missing arguments gracefully
1142 | 5. Consider versioning for prompt templates
1143 | 6. Cache dynamic content when appropriate
1144 | 7. Implement error handling
1145 | 8. Document expected argument formats
1146 | 9. Consider prompt composability
1147 | 10. Test prompts with various inputs
1148 |
1149 | ## UI integration
1150 |
1151 | Prompts can be surfaced in client UIs as:
1152 |
1153 | * Slash commands
1154 | * Quick actions
1155 | * Context menu items
1156 | * Command palette entries
1157 | * Guided workflows
1158 | * Interactive forms
1159 |
1160 | ## Updates and changes
1161 |
1162 | Servers can notify clients about prompt changes:
1163 |
1164 | 1. Server capability: `prompts.listChanged`
1165 | 2. Notification: `notifications/prompts/list_changed`
1166 | 3. Client re-fetches prompt list
1167 |
1168 | ## Security considerations
1169 |
1170 | When implementing prompts:
1171 |
1172 | * Validate all arguments
1173 | * Sanitize user input
1174 | * Consider rate limiting
1175 | * Implement access controls
1176 | * Audit prompt usage
1177 | * Handle sensitive data appropriately
1178 | * Validate generated content
1179 | * Implement timeouts
1180 | * Consider prompt injection risks
1181 | * Document security requirements
1182 |
1183 |
1184 | # Resources
1185 | Source: https://modelcontextprotocol.io/docs/concepts/resources
1186 |
1187 | Expose data and content from your servers to LLMs
1188 |
1189 | 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.
1190 |
1191 | <Note>
1192 | Resources are designed to be **application-controlled**, meaning that the client application can decide how and when they should be used.
1193 | Different MCP clients may handle resources differently. For example:
1194 |
1195 | * Claude Desktop currently requires users to explicitly select resources before they can be used
1196 | * Other clients might automatically select resources based on heuristics
1197 | * Some implementations may even allow the AI model itself to determine which resources to use
1198 |
1199 | 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).
1200 | </Note>
1201 |
1202 | ## Overview
1203 |
1204 | Resources represent any kind of data that an MCP server wants to make available to clients. This can include:
1205 |
1206 | * File contents
1207 | * Database records
1208 | * API responses
1209 | * Live system data
1210 | * Screenshots and images
1211 | * Log files
1212 | * And more
1213 |
1214 | Each resource is identified by a unique URI and can contain either text or binary data.
1215 |
1216 | ## Resource URIs
1217 |
1218 | Resources are identified using URIs that follow this format:
1219 |
1220 | ```
1221 | [protocol]://[host]/[path]
1222 | ```
1223 |
1224 | For example:
1225 |
1226 | * `file:///home/user/documents/report.pdf`
1227 | * `postgres://database/customers/schema`
1228 | * `screen://localhost/display1`
1229 |
1230 | The protocol and path structure is defined by the MCP server implementation. Servers can define their own custom URI schemes.
1231 |
1232 | ## Resource types
1233 |
1234 | Resources can contain two types of content:
1235 |
1236 | ### Text resources
1237 |
1238 | Text resources contain UTF-8 encoded text data. These are suitable for:
1239 |
1240 | * Source code
1241 | * Configuration files
1242 | * Log files
1243 | * JSON/XML data
1244 | * Plain text
1245 |
1246 | ### Binary resources
1247 |
1248 | Binary resources contain raw binary data encoded in base64. These are suitable for:
1249 |
1250 | * Images
1251 | * PDFs
1252 | * Audio files
1253 | * Video files
1254 | * Other non-text formats
1255 |
1256 | ## Resource discovery
1257 |
1258 | Clients can discover available resources through two main methods:
1259 |
1260 | ### Direct resources
1261 |
1262 | Servers expose a list of concrete resources via the `resources/list` endpoint. Each resource includes:
1263 |
1264 | ```typescript
1265 | {
1266 | uri: string; // Unique identifier for the resource
1267 | name: string; // Human-readable name
1268 | description?: string; // Optional description
1269 | mimeType?: string; // Optional MIME type
1270 | }
1271 | ```
1272 |
1273 | ### Resource templates
1274 |
1275 | For dynamic resources, servers can expose [URI templates](https://datatracker.ietf.org/doc/html/rfc6570) that clients can use to construct valid resource URIs:
1276 |
1277 | ```typescript
1278 | {
1279 | uriTemplate: string; // URI template following RFC 6570
1280 | name: string; // Human-readable name for this type
1281 | description?: string; // Optional description
1282 | mimeType?: string; // Optional MIME type for all matching resources
1283 | }
1284 | ```
1285 |
1286 | ## Reading resources
1287 |
1288 | To read a resource, clients make a `resources/read` request with the resource URI.
1289 |
1290 | The server responds with a list of resource contents:
1291 |
1292 | ```typescript
1293 | {
1294 | contents: [
1295 | {
1296 | uri: string; // The URI of the resource
1297 | mimeType?: string; // Optional MIME type
1298 |
1299 | // One of:
1300 | text?: string; // For text resources
1301 | blob?: string; // For binary resources (base64 encoded)
1302 | }
1303 | ]
1304 | }
1305 | ```
1306 |
1307 | <Tip>
1308 | 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.
1309 | </Tip>
1310 |
1311 | ## Resource updates
1312 |
1313 | MCP supports real-time updates for resources through two mechanisms:
1314 |
1315 | ### List changes
1316 |
1317 | Servers can notify clients when their list of available resources changes via the `notifications/resources/list_changed` notification.
1318 |
1319 | ### Content changes
1320 |
1321 | Clients can subscribe to updates for specific resources:
1322 |
1323 | 1. Client sends `resources/subscribe` with resource URI
1324 | 2. Server sends `notifications/resources/updated` when the resource changes
1325 | 3. Client can fetch latest content with `resources/read`
1326 | 4. Client can unsubscribe with `resources/unsubscribe`
1327 |
1328 | ## Example implementation
1329 |
1330 | Here's a simple example of implementing resource support in an MCP server:
1331 |
1332 | <Tabs>
1333 | <Tab title="TypeScript">
1334 | ```typescript
1335 | const server = new Server({
1336 | name: "example-server",
1337 | version: "1.0.0"
1338 | }, {
1339 | capabilities: {
1340 | resources: {}
1341 | }
1342 | });
1343 |
1344 | // List available resources
1345 | server.setRequestHandler(ListResourcesRequestSchema, async () => {
1346 | return {
1347 | resources: [
1348 | {
1349 | uri: "file:///logs/app.log",
1350 | name: "Application Logs",
1351 | mimeType: "text/plain"
1352 | }
1353 | ]
1354 | };
1355 | });
1356 |
1357 | // Read resource contents
1358 | server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
1359 | const uri = request.params.uri;
1360 |
1361 | if (uri === "file:///logs/app.log") {
1362 | const logContents = await readLogFile();
1363 | return {
1364 | contents: [
1365 | {
1366 | uri,
1367 | mimeType: "text/plain",
1368 | text: logContents
1369 | }
1370 | ]
1371 | };
1372 | }
1373 |
1374 | throw new Error("Resource not found");
1375 | });
1376 | ```
1377 | </Tab>
1378 |
1379 | <Tab title="Python">
1380 | ```python
1381 | app = Server("example-server")
1382 |
1383 | @app.list_resources()
1384 | async def list_resources() -> list[types.Resource]:
1385 | return [
1386 | types.Resource(
1387 | uri="file:///logs/app.log",
1388 | name="Application Logs",
1389 | mimeType="text/plain"
1390 | )
1391 | ]
1392 |
1393 | @app.read_resource()
1394 | async def read_resource(uri: AnyUrl) -> str:
1395 | if str(uri) == "file:///logs/app.log":
1396 | log_contents = await read_log_file()
1397 | return log_contents
1398 |
1399 | raise ValueError("Resource not found")
1400 |
1401 | # Start server
1402 | async with stdio_server() as streams:
1403 | await app.run(
1404 | streams[0],
1405 | streams[1],
1406 | app.create_initialization_options()
1407 | )
1408 | ```
1409 | </Tab>
1410 | </Tabs>
1411 |
1412 | ## Best practices
1413 |
1414 | When implementing resource support:
1415 |
1416 | 1. Use clear, descriptive resource names and URIs
1417 | 2. Include helpful descriptions to guide LLM understanding
1418 | 3. Set appropriate MIME types when known
1419 | 4. Implement resource templates for dynamic content
1420 | 5. Use subscriptions for frequently changing resources
1421 | 6. Handle errors gracefully with clear error messages
1422 | 7. Consider pagination for large resource lists
1423 | 8. Cache resource contents when appropriate
1424 | 9. Validate URIs before processing
1425 | 10. Document your custom URI schemes
1426 |
1427 | ## Security considerations
1428 |
1429 | When exposing resources:
1430 |
1431 | * Validate all resource URIs
1432 | * Implement appropriate access controls
1433 | * Sanitize file paths to prevent directory traversal
1434 | * Be cautious with binary data handling
1435 | * Consider rate limiting for resource reads
1436 | * Audit resource access
1437 | * Encrypt sensitive data in transit
1438 | * Validate MIME types
1439 | * Implement timeouts for long-running reads
1440 | * Handle resource cleanup appropriately
1441 |
1442 |
1443 | # Roots
1444 | Source: https://modelcontextprotocol.io/docs/concepts/roots
1445 |
1446 | Understanding roots in MCP
1447 |
1448 | 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.
1449 |
1450 | ## What are Roots?
1451 |
1452 | 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.
1453 |
1454 | For example, roots could be:
1455 |
1456 | ```
1457 | file:///home/user/projects/myapp
1458 | https://api.example.com/v1
1459 | ```
1460 |
1461 | ## Why Use Roots?
1462 |
1463 | Roots serve several important purposes:
1464 |
1465 | 1. **Guidance**: They inform servers about relevant resources and locations
1466 | 2. **Clarity**: Roots make it clear which resources are part of your workspace
1467 | 3. **Organization**: Multiple roots let you work with different resources simultaneously
1468 |
1469 | ## How Roots Work
1470 |
1471 | When a client supports roots, it:
1472 |
1473 | 1. Declares the `roots` capability during connection
1474 | 2. Provides a list of suggested roots to the server
1475 | 3. Notifies the server when roots change (if supported)
1476 |
1477 | While roots are informational and not strictly enforcing, servers should:
1478 |
1479 | 1. Respect the provided roots
1480 | 2. Use root URIs to locate and access resources
1481 | 3. Prioritize operations within root boundaries
1482 |
1483 | ## Common Use Cases
1484 |
1485 | Roots are commonly used to define:
1486 |
1487 | * Project directories
1488 | * Repository locations
1489 | * API endpoints
1490 | * Configuration locations
1491 | * Resource boundaries
1492 |
1493 | ## Best Practices
1494 |
1495 | When working with roots:
1496 |
1497 | 1. Only suggest necessary resources
1498 | 2. Use clear, descriptive names for roots
1499 | 3. Monitor root accessibility
1500 | 4. Handle root changes gracefully
1501 |
1502 | ## Example
1503 |
1504 | Here's how a typical MCP client might expose roots:
1505 |
1506 | ```json
1507 | {
1508 | "roots": [
1509 | {
1510 | "uri": "file:///home/user/projects/frontend",
1511 | "name": "Frontend Repository"
1512 | },
1513 | {
1514 | "uri": "https://api.example.com/v1",
1515 | "name": "API Endpoint"
1516 | }
1517 | ]
1518 | }
1519 | ```
1520 |
1521 | This configuration suggests the server focus on both a local repository and an API endpoint while keeping them logically separated.
1522 |
1523 |
1524 | # Sampling
1525 | Source: https://modelcontextprotocol.io/docs/concepts/sampling
1526 |
1527 | Let your servers request completions from LLMs
1528 |
1529 | 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.
1530 |
1531 | <Info>
1532 | This feature of MCP is not yet supported in the Claude Desktop client.
1533 | </Info>
1534 |
1535 | ## How sampling works
1536 |
1537 | The sampling flow follows these steps:
1538 |
1539 | 1. Server sends a `sampling/createMessage` request to the client
1540 | 2. Client reviews the request and can modify it
1541 | 3. Client samples from an LLM
1542 | 4. Client reviews the completion
1543 | 5. Client returns the result to the server
1544 |
1545 | This human-in-the-loop design ensures users maintain control over what the LLM sees and generates.
1546 |
1547 | ## Message format
1548 |
1549 | Sampling requests use a standardized message format:
1550 |
1551 | ```typescript
1552 | {
1553 | messages: [
1554 | {
1555 | role: "user" | "assistant",
1556 | content: {
1557 | type: "text" | "image",
1558 |
1559 | // For text:
1560 | text?: string,
1561 |
1562 | // For images:
1563 | data?: string, // base64 encoded
1564 | mimeType?: string
1565 | }
1566 | }
1567 | ],
1568 | modelPreferences?: {
1569 | hints?: [{
1570 | name?: string // Suggested model name/family
1571 | }],
1572 | costPriority?: number, // 0-1, importance of minimizing cost
1573 | speedPriority?: number, // 0-1, importance of low latency
1574 | intelligencePriority?: number // 0-1, importance of capabilities
1575 | },
1576 | systemPrompt?: string,
1577 | includeContext?: "none" | "thisServer" | "allServers",
1578 | temperature?: number,
1579 | maxTokens: number,
1580 | stopSequences?: string[],
1581 | metadata?: Record<string, unknown>
1582 | }
1583 | ```
1584 |
1585 | ## Request parameters
1586 |
1587 | ### Messages
1588 |
1589 | The `messages` array contains the conversation history to send to the LLM. Each message has:
1590 |
1591 | * `role`: Either "user" or "assistant"
1592 | * `content`: The message content, which can be:
1593 | * Text content with a `text` field
1594 | * Image content with `data` (base64) and `mimeType` fields
1595 |
1596 | ### Model preferences
1597 |
1598 | The `modelPreferences` object allows servers to specify their model selection preferences:
1599 |
1600 | * `hints`: Array of model name suggestions that clients can use to select an appropriate model:
1601 | * `name`: String that can match full or partial model names (e.g. "claude-3", "sonnet")
1602 | * Clients may map hints to equivalent models from different providers
1603 | * Multiple hints are evaluated in preference order
1604 |
1605 | * Priority values (0-1 normalized):
1606 | * `costPriority`: Importance of minimizing costs
1607 | * `speedPriority`: Importance of low latency response
1608 | * `intelligencePriority`: Importance of advanced model capabilities
1609 |
1610 | Clients make the final model selection based on these preferences and their available models.
1611 |
1612 | ### System prompt
1613 |
1614 | An optional `systemPrompt` field allows servers to request a specific system prompt. The client may modify or ignore this.
1615 |
1616 | ### Context inclusion
1617 |
1618 | The `includeContext` parameter specifies what MCP context to include:
1619 |
1620 | * `"none"`: No additional context
1621 | * `"thisServer"`: Include context from the requesting server
1622 | * `"allServers"`: Include context from all connected MCP servers
1623 |
1624 | The client controls what context is actually included.
1625 |
1626 | ### Sampling parameters
1627 |
1628 | Fine-tune the LLM sampling with:
1629 |
1630 | * `temperature`: Controls randomness (0.0 to 1.0)
1631 | * `maxTokens`: Maximum tokens to generate
1632 | * `stopSequences`: Array of sequences that stop generation
1633 | * `metadata`: Additional provider-specific parameters
1634 |
1635 | ## Response format
1636 |
1637 | The client returns a completion result:
1638 |
1639 | ```typescript
1640 | {
1641 | model: string, // Name of the model used
1642 | stopReason?: "endTurn" | "stopSequence" | "maxTokens" | string,
1643 | role: "user" | "assistant",
1644 | content: {
1645 | type: "text" | "image",
1646 | text?: string,
1647 | data?: string,
1648 | mimeType?: string
1649 | }
1650 | }
1651 | ```
1652 |
1653 | ## Example request
1654 |
1655 | Here's an example of requesting sampling from a client:
1656 |
1657 | ```json
1658 | {
1659 | "method": "sampling/createMessage",
1660 | "params": {
1661 | "messages": [
1662 | {
1663 | "role": "user",
1664 | "content": {
1665 | "type": "text",
1666 | "text": "What files are in the current directory?"
1667 | }
1668 | }
1669 | ],
1670 | "systemPrompt": "You are a helpful file system assistant.",
1671 | "includeContext": "thisServer",
1672 | "maxTokens": 100
1673 | }
1674 | }
1675 | ```
1676 |
1677 | ## Best practices
1678 |
1679 | When implementing sampling:
1680 |
1681 | 1. Always provide clear, well-structured prompts
1682 | 2. Handle both text and image content appropriately
1683 | 3. Set reasonable token limits
1684 | 4. Include relevant context through `includeContext`
1685 | 5. Validate responses before using them
1686 | 6. Handle errors gracefully
1687 | 7. Consider rate limiting sampling requests
1688 | 8. Document expected sampling behavior
1689 | 9. Test with various model parameters
1690 | 10. Monitor sampling costs
1691 |
1692 | ## Human in the loop controls
1693 |
1694 | Sampling is designed with human oversight in mind:
1695 |
1696 | ### For prompts
1697 |
1698 | * Clients should show users the proposed prompt
1699 | * Users should be able to modify or reject prompts
1700 | * System prompts can be filtered or modified
1701 | * Context inclusion is controlled by the client
1702 |
1703 | ### For completions
1704 |
1705 | * Clients should show users the completion
1706 | * Users should be able to modify or reject completions
1707 | * Clients can filter or modify completions
1708 | * Users control which model is used
1709 |
1710 | ## Security considerations
1711 |
1712 | When implementing sampling:
1713 |
1714 | * Validate all message content
1715 | * Sanitize sensitive information
1716 | * Implement appropriate rate limits
1717 | * Monitor sampling usage
1718 | * Encrypt data in transit
1719 | * Handle user data privacy
1720 | * Audit sampling requests
1721 | * Control cost exposure
1722 | * Implement timeouts
1723 | * Handle model errors gracefully
1724 |
1725 | ## Common patterns
1726 |
1727 | ### Agentic workflows
1728 |
1729 | Sampling enables agentic patterns like:
1730 |
1731 | * Reading and analyzing resources
1732 | * Making decisions based on context
1733 | * Generating structured data
1734 | * Handling multi-step tasks
1735 | * Providing interactive assistance
1736 |
1737 | ### Context management
1738 |
1739 | Best practices for context:
1740 |
1741 | * Request minimal necessary context
1742 | * Structure context clearly
1743 | * Handle context size limits
1744 | * Update context as needed
1745 | * Clean up stale context
1746 |
1747 | ### Error handling
1748 |
1749 | Robust error handling should:
1750 |
1751 | * Catch sampling failures
1752 | * Handle timeout errors
1753 | * Manage rate limits
1754 | * Validate responses
1755 | * Provide fallback behaviors
1756 | * Log errors appropriately
1757 |
1758 | ## Limitations
1759 |
1760 | Be aware of these limitations:
1761 |
1762 | * Sampling depends on client capabilities
1763 | * Users control sampling behavior
1764 | * Context size has limits
1765 | * Rate limits may apply
1766 | * Costs should be considered
1767 | * Model availability varies
1768 | * Response times vary
1769 | * Not all content types supported
1770 |
1771 |
1772 | # Tools
1773 | Source: https://modelcontextprotocol.io/docs/concepts/tools
1774 |
1775 | Enable LLMs to perform actions through your server
1776 |
1777 | 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.
1778 |
1779 | <Note>
1780 | 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).
1781 | </Note>
1782 |
1783 | ## Overview
1784 |
1785 | 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:
1786 |
1787 | * **Discovery**: Clients can list available tools through the `tools/list` endpoint
1788 | * **Invocation**: Tools are called using the `tools/call` endpoint, where servers perform the requested operation and return results
1789 | * **Flexibility**: Tools can range from simple calculations to complex API interactions
1790 |
1791 | 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.
1792 |
1793 | ## Tool definition structure
1794 |
1795 | Each tool is defined with the following structure:
1796 |
1797 | ```typescript
1798 | {
1799 | name: string; // Unique identifier for the tool
1800 | description?: string; // Human-readable description
1801 | inputSchema: { // JSON Schema for the tool's parameters
1802 | type: "object",
1803 | properties: { ... } // Tool-specific parameters
1804 | }
1805 | }
1806 | ```
1807 |
1808 | ## Implementing tools
1809 |
1810 | Here's an example of implementing a basic tool in an MCP server:
1811 |
1812 | <Tabs>
1813 | <Tab title="TypeScript">
1814 | ```typescript
1815 | const server = new Server({
1816 | name: "example-server",
1817 | version: "1.0.0"
1818 | }, {
1819 | capabilities: {
1820 | tools: {}
1821 | }
1822 | });
1823 |
1824 | // Define available tools
1825 | server.setRequestHandler(ListToolsRequestSchema, async () => {
1826 | return {
1827 | tools: [{
1828 | name: "calculate_sum",
1829 | description: "Add two numbers together",
1830 | inputSchema: {
1831 | type: "object",
1832 | properties: {
1833 | a: { type: "number" },
1834 | b: { type: "number" }
1835 | },
1836 | required: ["a", "b"]
1837 | }
1838 | }]
1839 | };
1840 | });
1841 |
1842 | // Handle tool execution
1843 | server.setRequestHandler(CallToolRequestSchema, async (request) => {
1844 | if (request.params.name === "calculate_sum") {
1845 | const { a, b } = request.params.arguments;
1846 | return {
1847 | content: [
1848 | {
1849 | type: "text",
1850 | text: String(a + b)
1851 | }
1852 | ]
1853 | };
1854 | }
1855 | throw new Error("Tool not found");
1856 | });
1857 | ```
1858 | </Tab>
1859 |
1860 | <Tab title="Python">
1861 | ```python
1862 | app = Server("example-server")
1863 |
1864 | @app.list_tools()
1865 | async def list_tools() -> list[types.Tool]:
1866 | return [
1867 | types.Tool(
1868 | name="calculate_sum",
1869 | description="Add two numbers together",
1870 | inputSchema={
1871 | "type": "object",
1872 | "properties": {
1873 | "a": {"type": "number"},
1874 | "b": {"type": "number"}
1875 | },
1876 | "required": ["a", "b"]
1877 | }
1878 | )
1879 | ]
1880 |
1881 | @app.call_tool()
1882 | async def call_tool(
1883 | name: str,
1884 | arguments: dict
1885 | ) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
1886 | if name == "calculate_sum":
1887 | a = arguments["a"]
1888 | b = arguments["b"]
1889 | result = a + b
1890 | return [types.TextContent(type="text", text=str(result))]
1891 | raise ValueError(f"Tool not found: {name}")
1892 | ```
1893 | </Tab>
1894 | </Tabs>
1895 |
1896 | ## Example tool patterns
1897 |
1898 | Here are some examples of types of tools that a server could provide:
1899 |
1900 | ### System operations
1901 |
1902 | Tools that interact with the local system:
1903 |
1904 | ```typescript
1905 | {
1906 | name: "execute_command",
1907 | description: "Run a shell command",
1908 | inputSchema: {
1909 | type: "object",
1910 | properties: {
1911 | command: { type: "string" },
1912 | args: { type: "array", items: { type: "string" } }
1913 | }
1914 | }
1915 | }
1916 | ```
1917 |
1918 | ### API integrations
1919 |
1920 | Tools that wrap external APIs:
1921 |
1922 | ```typescript
1923 | {
1924 | name: "github_create_issue",
1925 | description: "Create a GitHub issue",
1926 | inputSchema: {
1927 | type: "object",
1928 | properties: {
1929 | title: { type: "string" },
1930 | body: { type: "string" },
1931 | labels: { type: "array", items: { type: "string" } }
1932 | }
1933 | }
1934 | }
1935 | ```
1936 |
1937 | ### Data processing
1938 |
1939 | Tools that transform or analyze data:
1940 |
1941 | ```typescript
1942 | {
1943 | name: "analyze_csv",
1944 | description: "Analyze a CSV file",
1945 | inputSchema: {
1946 | type: "object",
1947 | properties: {
1948 | filepath: { type: "string" },
1949 | operations: {
1950 | type: "array",
1951 | items: {
1952 | enum: ["sum", "average", "count"]
1953 | }
1954 | }
1955 | }
1956 | }
1957 | }
1958 | ```
1959 |
1960 | ## Best practices
1961 |
1962 | When implementing tools:
1963 |
1964 | 1. Provide clear, descriptive names and descriptions
1965 | 2. Use detailed JSON Schema definitions for parameters
1966 | 3. Include examples in tool descriptions to demonstrate how the model should use them
1967 | 4. Implement proper error handling and validation
1968 | 5. Use progress reporting for long operations
1969 | 6. Keep tool operations focused and atomic
1970 | 7. Document expected return value structures
1971 | 8. Implement proper timeouts
1972 | 9. Consider rate limiting for resource-intensive operations
1973 | 10. Log tool usage for debugging and monitoring
1974 |
1975 | ## Security considerations
1976 |
1977 | When exposing tools:
1978 |
1979 | ### Input validation
1980 |
1981 | * Validate all parameters against the schema
1982 | * Sanitize file paths and system commands
1983 | * Validate URLs and external identifiers
1984 | * Check parameter sizes and ranges
1985 | * Prevent command injection
1986 |
1987 | ### Access control
1988 |
1989 | * Implement authentication where needed
1990 | * Use appropriate authorization checks
1991 | * Audit tool usage
1992 | * Rate limit requests
1993 | * Monitor for abuse
1994 |
1995 | ### Error handling
1996 |
1997 | * Don't expose internal errors to clients
1998 | * Log security-relevant errors
1999 | * Handle timeouts appropriately
2000 | * Clean up resources after errors
2001 | * Validate return values
2002 |
2003 | ## Tool discovery and updates
2004 |
2005 | MCP supports dynamic tool discovery:
2006 |
2007 | 1. Clients can list available tools at any time
2008 | 2. Servers can notify clients when tools change using `notifications/tools/list_changed`
2009 | 3. Tools can be added or removed during runtime
2010 | 4. Tool definitions can be updated (though this should be done carefully)
2011 |
2012 | ## Error handling
2013 |
2014 | 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:
2015 |
2016 | 1. Set `isError` to `true` in the result
2017 | 2. Include error details in the `content` array
2018 |
2019 | Here's an example of proper error handling for tools:
2020 |
2021 | <Tabs>
2022 | <Tab title="TypeScript">
2023 | ```typescript
2024 | try {
2025 | // Tool operation
2026 | const result = performOperation();
2027 | return {
2028 | content: [
2029 | {
2030 | type: "text",
2031 | text: `Operation successful: ${result}`
2032 | }
2033 | ]
2034 | };
2035 | } catch (error) {
2036 | return {
2037 | isError: true,
2038 | content: [
2039 | {
2040 | type: "text",
2041 | text: `Error: ${error.message}`
2042 | }
2043 | ]
2044 | };
2045 | }
2046 | ```
2047 | </Tab>
2048 |
2049 | <Tab title="Python">
2050 | ```python
2051 | try:
2052 | # Tool operation
2053 | result = perform_operation()
2054 | return types.CallToolResult(
2055 | content=[
2056 | types.TextContent(
2057 | type="text",
2058 | text=f"Operation successful: {result}"
2059 | )
2060 | ]
2061 | )
2062 | except Exception as error:
2063 | return types.CallToolResult(
2064 | isError=True,
2065 | content=[
2066 | types.TextContent(
2067 | type="text",
2068 | text=f"Error: {str(error)}"
2069 | )
2070 | ]
2071 | )
2072 | ```
2073 | </Tab>
2074 | </Tabs>
2075 |
2076 | This approach allows the LLM to see that an error occurred and potentially take corrective action or request human intervention.
2077 |
2078 | ## Testing tools
2079 |
2080 | A comprehensive testing strategy for MCP tools should cover:
2081 |
2082 | * **Functional testing**: Verify tools execute correctly with valid inputs and handle invalid inputs appropriately
2083 | * **Integration testing**: Test tool interaction with external systems using both real and mocked dependencies
2084 | * **Security testing**: Validate authentication, authorization, input sanitization, and rate limiting
2085 | * **Performance testing**: Check behavior under load, timeout handling, and resource cleanup
2086 | * **Error handling**: Ensure tools properly report errors through the MCP protocol and clean up resources
2087 |
2088 |
2089 | # Transports
2090 | Source: https://modelcontextprotocol.io/docs/concepts/transports
2091 |
2092 | Learn about MCP's communication mechanisms
2093 |
2094 | 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.
2095 |
2096 | ## Message Format
2097 |
2098 | 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.
2099 |
2100 | There are three types of JSON-RPC messages used:
2101 |
2102 | ### Requests
2103 |
2104 | ```typescript
2105 | {
2106 | jsonrpc: "2.0",
2107 | id: number | string,
2108 | method: string,
2109 | params?: object
2110 | }
2111 | ```
2112 |
2113 | ### Responses
2114 |
2115 | ```typescript
2116 | {
2117 | jsonrpc: "2.0",
2118 | id: number | string,
2119 | result?: object,
2120 | error?: {
2121 | code: number,
2122 | message: string,
2123 | data?: unknown
2124 | }
2125 | }
2126 | ```
2127 |
2128 | ### Notifications
2129 |
2130 | ```typescript
2131 | {
2132 | jsonrpc: "2.0",
2133 | method: string,
2134 | params?: object
2135 | }
2136 | ```
2137 |
2138 | ## Built-in Transport Types
2139 |
2140 | MCP includes two standard transport implementations:
2141 |
2142 | ### Standard Input/Output (stdio)
2143 |
2144 | The stdio transport enables communication through standard input and output streams. This is particularly useful for local integrations and command-line tools.
2145 |
2146 | Use stdio when:
2147 |
2148 | * Building command-line tools
2149 | * Implementing local integrations
2150 | * Needing simple process communication
2151 | * Working with shell scripts
2152 |
2153 | <Tabs>
2154 | <Tab title="TypeScript (Server)">
2155 | ```typescript
2156 | const server = new Server({
2157 | name: "example-server",
2158 | version: "1.0.0"
2159 | }, {
2160 | capabilities: {}
2161 | });
2162 |
2163 | const transport = new StdioServerTransport();
2164 | await server.connect(transport);
2165 | ```
2166 | </Tab>
2167 |
2168 | <Tab title="TypeScript (Client)">
2169 | ```typescript
2170 | const client = new Client({
2171 | name: "example-client",
2172 | version: "1.0.0"
2173 | }, {
2174 | capabilities: {}
2175 | });
2176 |
2177 | const transport = new StdioClientTransport({
2178 | command: "./server",
2179 | args: ["--option", "value"]
2180 | });
2181 | await client.connect(transport);
2182 | ```
2183 | </Tab>
2184 |
2185 | <Tab title="Python (Server)">
2186 | ```python
2187 | app = Server("example-server")
2188 |
2189 | async with stdio_server() as streams:
2190 | await app.run(
2191 | streams[0],
2192 | streams[1],
2193 | app.create_initialization_options()
2194 | )
2195 | ```
2196 | </Tab>
2197 |
2198 | <Tab title="Python (Client)">
2199 | ```python
2200 | params = StdioServerParameters(
2201 | command="./server",
2202 | args=["--option", "value"]
2203 | )
2204 |
2205 | async with stdio_client(params) as streams:
2206 | async with ClientSession(streams[0], streams[1]) as session:
2207 | await session.initialize()
2208 | ```
2209 | </Tab>
2210 | </Tabs>
2211 |
2212 | ### Server-Sent Events (SSE)
2213 |
2214 | SSE transport enables server-to-client streaming with HTTP POST requests for client-to-server communication.
2215 |
2216 | Use SSE when:
2217 |
2218 | * Only server-to-client streaming is needed
2219 | * Working with restricted networks
2220 | * Implementing simple updates
2221 |
2222 | <Tabs>
2223 | <Tab title="TypeScript (Server)">
2224 | ```typescript
2225 | import express from "express";
2226 |
2227 | const app = express();
2228 |
2229 | const server = new Server({
2230 | name: "example-server",
2231 | version: "1.0.0"
2232 | }, {
2233 | capabilities: {}
2234 | });
2235 |
2236 | let transport: SSEServerTransport | null = null;
2237 |
2238 | app.get("/sse", (req, res) => {
2239 | transport = new SSEServerTransport("/messages", res);
2240 | server.connect(transport);
2241 | });
2242 |
2243 | app.post("/messages", (req, res) => {
2244 | if (transport) {
2245 | transport.handlePostMessage(req, res);
2246 | }
2247 | });
2248 |
2249 | app.listen(3000);
2250 | ```
2251 | </Tab>
2252 |
2253 | <Tab title="TypeScript (Client)">
2254 | ```typescript
2255 | const client = new Client({
2256 | name: "example-client",
2257 | version: "1.0.0"
2258 | }, {
2259 | capabilities: {}
2260 | });
2261 |
2262 | const transport = new SSEClientTransport(
2263 | new URL("http://localhost:3000/sse")
2264 | );
2265 | await client.connect(transport);
2266 | ```
2267 | </Tab>
2268 |
2269 | <Tab title="Python (Server)">
2270 | ```python
2271 | from mcp.server.sse import SseServerTransport
2272 | from starlette.applications import Starlette
2273 | from starlette.routing import Route
2274 |
2275 | app = Server("example-server")
2276 | sse = SseServerTransport("/messages")
2277 |
2278 | async def handle_sse(scope, receive, send):
2279 | async with sse.connect_sse(scope, receive, send) as streams:
2280 | await app.run(streams[0], streams[1], app.create_initialization_options())
2281 |
2282 | async def handle_messages(scope, receive, send):
2283 | await sse.handle_post_message(scope, receive, send)
2284 |
2285 | starlette_app = Starlette(
2286 | routes=[
2287 | Route("/sse", endpoint=handle_sse),
2288 | Route("/messages", endpoint=handle_messages, methods=["POST"]),
2289 | ]
2290 | )
2291 | ```
2292 | </Tab>
2293 |
2294 | <Tab title="Python (Client)">
2295 | ```python
2296 | async with sse_client("http://localhost:8000/sse") as streams:
2297 | async with ClientSession(streams[0], streams[1]) as session:
2298 | await session.initialize()
2299 | ```
2300 | </Tab>
2301 | </Tabs>
2302 |
2303 | ## Custom Transports
2304 |
2305 | MCP makes it easy to implement custom transports for specific needs. Any transport implementation just needs to conform to the Transport interface:
2306 |
2307 | You can implement custom transports for:
2308 |
2309 | * Custom network protocols
2310 | * Specialized communication channels
2311 | * Integration with existing systems
2312 | * Performance optimization
2313 |
2314 | <Tabs>
2315 | <Tab title="TypeScript">
2316 | ```typescript
2317 | interface Transport {
2318 | // Start processing messages
2319 | start(): Promise<void>;
2320 |
2321 | // Send a JSON-RPC message
2322 | send(message: JSONRPCMessage): Promise<void>;
2323 |
2324 | // Close the connection
2325 | close(): Promise<void>;
2326 |
2327 | // Callbacks
2328 | onclose?: () => void;
2329 | onerror?: (error: Error) => void;
2330 | onmessage?: (message: JSONRPCMessage) => void;
2331 | }
2332 | ```
2333 | </Tab>
2334 |
2335 | <Tab title="Python">
2336 | Note that while MCP Servers are often implemented with asyncio, we recommend
2337 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2338 |
2339 | ```python
2340 | @contextmanager
2341 | async def create_transport(
2342 | read_stream: MemoryObjectReceiveStream[JSONRPCMessage | Exception],
2343 | write_stream: MemoryObjectSendStream[JSONRPCMessage]
2344 | ):
2345 | """
2346 | Transport interface for MCP.
2347 |
2348 | Args:
2349 | read_stream: Stream to read incoming messages from
2350 | write_stream: Stream to write outgoing messages to
2351 | """
2352 | async with anyio.create_task_group() as tg:
2353 | try:
2354 | # Start processing messages
2355 | tg.start_soon(lambda: process_messages(read_stream))
2356 |
2357 | # Send messages
2358 | async with write_stream:
2359 | yield write_stream
2360 |
2361 | except Exception as exc:
2362 | # Handle errors
2363 | raise exc
2364 | finally:
2365 | # Clean up
2366 | tg.cancel_scope.cancel()
2367 | await write_stream.aclose()
2368 | await read_stream.aclose()
2369 | ```
2370 | </Tab>
2371 | </Tabs>
2372 |
2373 | ## Error Handling
2374 |
2375 | Transport implementations should handle various error scenarios:
2376 |
2377 | 1. Connection errors
2378 | 2. Message parsing errors
2379 | 3. Protocol errors
2380 | 4. Network timeouts
2381 | 5. Resource cleanup
2382 |
2383 | Example error handling:
2384 |
2385 | <Tabs>
2386 | <Tab title="TypeScript">
2387 | ```typescript
2388 | class ExampleTransport implements Transport {
2389 | async start() {
2390 | try {
2391 | // Connection logic
2392 | } catch (error) {
2393 | this.onerror?.(new Error(`Failed to connect: ${error}`));
2394 | throw error;
2395 | }
2396 | }
2397 |
2398 | async send(message: JSONRPCMessage) {
2399 | try {
2400 | // Sending logic
2401 | } catch (error) {
2402 | this.onerror?.(new Error(`Failed to send message: ${error}`));
2403 | throw error;
2404 | }
2405 | }
2406 | }
2407 | ```
2408 | </Tab>
2409 |
2410 | <Tab title="Python">
2411 | Note that while MCP Servers are often implemented with asyncio, we recommend
2412 | implementing low-level interfaces like transports with `anyio` for wider compatibility.
2413 |
2414 | ```python
2415 | @contextmanager
2416 | async def example_transport(scope: Scope, receive: Receive, send: Send):
2417 | try:
2418 | # Create streams for bidirectional communication
2419 | read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
2420 | write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
2421 |
2422 | async def message_handler():
2423 | try:
2424 | async with read_stream_writer:
2425 | # Message handling logic
2426 | pass
2427 | except Exception as exc:
2428 | logger.error(f"Failed to handle message: {exc}")
2429 | raise exc
2430 |
2431 | async with anyio.create_task_group() as tg:
2432 | tg.start_soon(message_handler)
2433 | try:
2434 | # Yield streams for communication
2435 | yield read_stream, write_stream
2436 | except Exception as exc:
2437 | logger.error(f"Transport error: {exc}")
2438 | raise exc
2439 | finally:
2440 | tg.cancel_scope.cancel()
2441 | await write_stream.aclose()
2442 | await read_stream.aclose()
2443 | except Exception as exc:
2444 | logger.error(f"Failed to initialize transport: {exc}")
2445 | raise exc
2446 | ```
2447 | </Tab>
2448 | </Tabs>
2449 |
2450 | ## Best Practices
2451 |
2452 | When implementing or using MCP transport:
2453 |
2454 | 1. Handle connection lifecycle properly
2455 | 2. Implement proper error handling
2456 | 3. Clean up resources on connection close
2457 | 4. Use appropriate timeouts
2458 | 5. Validate messages before sending
2459 | 6. Log transport events for debugging
2460 | 7. Implement reconnection logic when appropriate
2461 | 8. Handle backpressure in message queues
2462 | 9. Monitor connection health
2463 | 10. Implement proper security measures
2464 |
2465 | ## Security Considerations
2466 |
2467 | When implementing transport:
2468 |
2469 | ### Authentication and Authorization
2470 |
2471 | * Implement proper authentication mechanisms
2472 | * Validate client credentials
2473 | * Use secure token handling
2474 | * Implement authorization checks
2475 |
2476 | ### Data Security
2477 |
2478 | * Use TLS for network transport
2479 | * Encrypt sensitive data
2480 | * Validate message integrity
2481 | * Implement message size limits
2482 | * Sanitize input data
2483 |
2484 | ### Network Security
2485 |
2486 | * Implement rate limiting
2487 | * Use appropriate timeouts
2488 | * Handle denial of service scenarios
2489 | * Monitor for unusual patterns
2490 | * Implement proper firewall rules
2491 |
2492 | ## Debugging Transport
2493 |
2494 | Tips for debugging transport issues:
2495 |
2496 | 1. Enable debug logging
2497 | 2. Monitor message flow
2498 | 3. Check connection states
2499 | 4. Validate message formats
2500 | 5. Test error scenarios
2501 | 6. Use network analysis tools
2502 | 7. Implement health checks
2503 | 8. Monitor resource usage
2504 | 9. Test edge cases
2505 | 10. Use proper error tracking
2506 |
2507 |
2508 | # Debugging
2509 | Source: https://modelcontextprotocol.io/docs/tools/debugging
2510 |
2511 | A comprehensive guide to debugging Model Context Protocol (MCP) integrations
2512 |
2513 | 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.
2514 |
2515 | <Info>
2516 | This guide is for macOS. Guides for other platforms are coming soon.
2517 | </Info>
2518 |
2519 | ## Debugging tools overview
2520 |
2521 | MCP provides several tools for debugging at different levels:
2522 |
2523 | 1. **MCP Inspector**
2524 | * Interactive debugging interface
2525 | * Direct server testing
2526 | * See the [Inspector guide](/docs/tools/inspector) for details
2527 |
2528 | 2. **Claude Desktop Developer Tools**
2529 | * Integration testing
2530 | * Log collection
2531 | * Chrome DevTools integration
2532 |
2533 | 3. **Server Logging**
2534 | * Custom logging implementations
2535 | * Error tracking
2536 | * Performance monitoring
2537 |
2538 | ## Debugging in Claude Desktop
2539 |
2540 | ### Checking server status
2541 |
2542 | The Claude.app interface provides basic server status information:
2543 |
2544 | 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:
2545 | * Connected servers
2546 | * Available prompts and resources
2547 |
2548 | 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:
2549 | * Tools made available to the model
2550 |
2551 | ### Viewing logs
2552 |
2553 | Review detailed MCP logs from Claude Desktop:
2554 |
2555 | ```bash
2556 | # Follow logs in real-time
2557 | tail -n 20 -F ~/Library/Logs/Claude/mcp*.log
2558 | ```
2559 |
2560 | The logs capture:
2561 |
2562 | * Server connection events
2563 | * Configuration issues
2564 | * Runtime errors
2565 | * Message exchanges
2566 |
2567 | ### Using Chrome DevTools
2568 |
2569 | Access Chrome's developer tools inside Claude Desktop to investigate client-side errors:
2570 |
2571 | 1. Create a `developer_settings.json` file with `allowDevTools` set to true:
2572 |
2573 | ```bash
2574 | echo '{"allowDevTools": true}' > ~/Library/Application\ Support/Claude/developer_settings.json
2575 | ```
2576 |
2577 | 2. Open DevTools: `Command-Option-Shift-i`
2578 |
2579 | Note: You'll see two DevTools windows:
2580 |
2581 | * Main content window
2582 | * App title bar window
2583 |
2584 | Use the Console panel to inspect client-side errors.
2585 |
2586 | Use the Network panel to inspect:
2587 |
2588 | * Message payloads
2589 | * Connection timing
2590 |
2591 | ## Common issues
2592 |
2593 | ### Working directory
2594 |
2595 | When using MCP servers with Claude Desktop:
2596 |
2597 | * 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
2598 | * Always use absolute paths in your configuration and `.env` files to ensure reliable operation
2599 | * For testing servers directly via command line, the working directory will be where you run the command
2600 |
2601 | For example in `claude_desktop_config.json`, use:
2602 |
2603 | ```json
2604 | {
2605 | "command": "npx",
2606 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/username/data"]
2607 | }
2608 | ```
2609 |
2610 | Instead of relative paths like `./data`
2611 |
2612 | ### Environment variables
2613 |
2614 | MCP servers inherit only a subset of environment variables automatically, like `USER`, `HOME`, and `PATH`.
2615 |
2616 | To override the default variables or provide your own, you can specify an `env` key in `claude_desktop_config.json`:
2617 |
2618 | ```json
2619 | {
2620 | "myserver": {
2621 | "command": "mcp-server-myapp",
2622 | "env": {
2623 | "MYAPP_API_KEY": "some_key",
2624 | }
2625 | }
2626 | }
2627 | ```
2628 |
2629 | ### Server initialization
2630 |
2631 | Common initialization problems:
2632 |
2633 | 1. **Path Issues**
2634 | * Incorrect server executable path
2635 | * Missing required files
2636 | * Permission problems
2637 | * Try using an absolute path for `command`
2638 |
2639 | 2. **Configuration Errors**
2640 | * Invalid JSON syntax
2641 | * Missing required fields
2642 | * Type mismatches
2643 |
2644 | 3. **Environment Problems**
2645 | * Missing environment variables
2646 | * Incorrect variable values
2647 | * Permission restrictions
2648 |
2649 | ### Connection problems
2650 |
2651 | When servers fail to connect:
2652 |
2653 | 1. Check Claude Desktop logs
2654 | 2. Verify server process is running
2655 | 3. Test standalone with [Inspector](/docs/tools/inspector)
2656 | 4. Verify protocol compatibility
2657 |
2658 | ## Implementing logging
2659 |
2660 | ### Server-side logging
2661 |
2662 | 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.
2663 |
2664 | <Warning>
2665 | Local MCP servers should not log messages to stdout (standard out), as this will interfere with protocol operation.
2666 | </Warning>
2667 |
2668 | For all [transports](/docs/concepts/transports), you can also provide logging to the client by sending a log message notification:
2669 |
2670 | <Tabs>
2671 | <Tab title="Python">
2672 | ```python
2673 | server.request_context.session.send_log_message(
2674 | level="info",
2675 | data="Server started successfully",
2676 | )
2677 | ```
2678 | </Tab>
2679 |
2680 | <Tab title="TypeScript">
2681 | ```typescript
2682 | server.sendLoggingMessage({
2683 | level: "info",
2684 | data: "Server started successfully",
2685 | });
2686 | ```
2687 | </Tab>
2688 | </Tabs>
2689 |
2690 | Important events to log:
2691 |
2692 | * Initialization steps
2693 | * Resource access
2694 | * Tool execution
2695 | * Error conditions
2696 | * Performance metrics
2697 |
2698 | ### Client-side logging
2699 |
2700 | In client applications:
2701 |
2702 | 1. Enable debug logging
2703 | 2. Monitor network traffic
2704 | 3. Track message exchanges
2705 | 4. Record error states
2706 |
2707 | ## Debugging workflow
2708 |
2709 | ### Development cycle
2710 |
2711 | 1. Initial Development
2712 | * Use [Inspector](/docs/tools/inspector) for basic testing
2713 | * Implement core functionality
2714 | * Add logging points
2715 |
2716 | 2. Integration Testing
2717 | * Test in Claude Desktop
2718 | * Monitor logs
2719 | * Check error handling
2720 |
2721 | ### Testing changes
2722 |
2723 | To test changes efficiently:
2724 |
2725 | * **Configuration changes**: Restart Claude Desktop
2726 | * **Server code changes**: Use Command-R to reload
2727 | * **Quick iteration**: Use [Inspector](/docs/tools/inspector) during development
2728 |
2729 | ## Best practices
2730 |
2731 | ### Logging strategy
2732 |
2733 | 1. **Structured Logging**
2734 | * Use consistent formats
2735 | * Include context
2736 | * Add timestamps
2737 | * Track request IDs
2738 |
2739 | 2. **Error Handling**
2740 | * Log stack traces
2741 | * Include error context
2742 | * Track error patterns
2743 | * Monitor recovery
2744 |
2745 | 3. **Performance Tracking**
2746 | * Log operation timing
2747 | * Monitor resource usage
2748 | * Track message sizes
2749 | * Measure latency
2750 |
2751 | ### Security considerations
2752 |
2753 | When debugging:
2754 |
2755 | 1. **Sensitive Data**
2756 | * Sanitize logs
2757 | * Protect credentials
2758 | * Mask personal information
2759 |
2760 | 2. **Access Control**
2761 | * Verify permissions
2762 | * Check authentication
2763 | * Monitor access patterns
2764 |
2765 | ## Getting help
2766 |
2767 | When encountering issues:
2768 |
2769 | 1. **First Steps**
2770 | * Check server logs
2771 | * Test with [Inspector](/docs/tools/inspector)
2772 | * Review configuration
2773 | * Verify environment
2774 |
2775 | 2. **Support Channels**
2776 | * GitHub issues
2777 | * GitHub discussions
2778 |
2779 | 3. **Providing Information**
2780 | * Log excerpts
2781 | * Configuration files
2782 | * Steps to reproduce
2783 | * Environment details
2784 |
2785 | ## Next steps
2786 |
2787 | <CardGroup cols={2}>
2788 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
2789 | Learn to use the MCP Inspector
2790 | </Card>
2791 | </CardGroup>
2792 |
2793 |
2794 | # Inspector
2795 | Source: https://modelcontextprotocol.io/docs/tools/inspector
2796 |
2797 | In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers
2798 |
2799 | 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.
2800 |
2801 | ## Getting started
2802 |
2803 | ### Installation and basic usage
2804 |
2805 | The Inspector runs directly through `npx` without requiring installation:
2806 |
2807 | ```bash
2808 | npx @modelcontextprotocol/inspector <command>
2809 | ```
2810 |
2811 | ```bash
2812 | npx @modelcontextprotocol/inspector <command> <arg1> <arg2>
2813 | ```
2814 |
2815 | #### Inspecting servers from NPM or PyPi
2816 |
2817 | A common way to start server packages from [NPM](https://npmjs.com) or [PyPi](https://pypi.com).
2818 |
2819 | <Tabs>
2820 | <Tab title="NPM package">
2821 | ```bash
2822 | npx -y @modelcontextprotocol/inspector npx <package-name> <args>
2823 | # For example
2824 | npx -y @modelcontextprotocol/inspector npx server-postgres postgres://127.0.0.1/testdb
2825 | ```
2826 | </Tab>
2827 |
2828 | <Tab title="PyPi package">
2829 | ```bash
2830 | npx @modelcontextprotocol/inspector uvx <package-name> <args>
2831 | # For example
2832 | npx @modelcontextprotocol/inspector uvx mcp-server-git --repository ~/code/mcp/servers.git
2833 | ```
2834 | </Tab>
2835 | </Tabs>
2836 |
2837 | #### Inspecting locally developed servers
2838 |
2839 | To inspect servers locally developed or downloaded as a repository, the most common
2840 | way is:
2841 |
2842 | <Tabs>
2843 | <Tab title="TypeScript">
2844 | ```bash
2845 | npx @modelcontextprotocol/inspector node path/to/server/index.js args...
2846 | ```
2847 | </Tab>
2848 |
2849 | <Tab title="Python">
2850 | ```bash
2851 | npx @modelcontextprotocol/inspector \
2852 | uv \
2853 | --directory path/to/server \
2854 | run \
2855 | package-name \
2856 | args...
2857 | ```
2858 | </Tab>
2859 | </Tabs>
2860 |
2861 | Please carefully read any attached README for the most accurate instructions.
2862 |
2863 | ## Feature overview
2864 |
2865 | <Frame caption="The MCP Inspector interface">
2866 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/mcp-inspector.png" />
2867 | </Frame>
2868 |
2869 | The Inspector provides several features for interacting with your MCP server:
2870 |
2871 | ### Server connection pane
2872 |
2873 | * Allows selecting the [transport](/docs/concepts/transports) for connecting to the server
2874 | * For local servers, supports customizing the command-line arguments and environment
2875 |
2876 | ### Resources tab
2877 |
2878 | * Lists all available resources
2879 | * Shows resource metadata (MIME types, descriptions)
2880 | * Allows resource content inspection
2881 | * Supports subscription testing
2882 |
2883 | ### Prompts tab
2884 |
2885 | * Displays available prompt templates
2886 | * Shows prompt arguments and descriptions
2887 | * Enables prompt testing with custom arguments
2888 | * Previews generated messages
2889 |
2890 | ### Tools tab
2891 |
2892 | * Lists available tools
2893 | * Shows tool schemas and descriptions
2894 | * Enables tool testing with custom inputs
2895 | * Displays tool execution results
2896 |
2897 | ### Notifications pane
2898 |
2899 | * Presents all logs recorded from the server
2900 | * Shows notifications received from the server
2901 |
2902 | ## Best practices
2903 |
2904 | ### Development workflow
2905 |
2906 | 1. Start Development
2907 | * Launch Inspector with your server
2908 | * Verify basic connectivity
2909 | * Check capability negotiation
2910 |
2911 | 2. Iterative testing
2912 | * Make server changes
2913 | * Rebuild the server
2914 | * Reconnect the Inspector
2915 | * Test affected features
2916 | * Monitor messages
2917 |
2918 | 3. Test edge cases
2919 | * Invalid inputs
2920 | * Missing prompt arguments
2921 | * Concurrent operations
2922 | * Verify error handling and error responses
2923 |
2924 | ## Next steps
2925 |
2926 | <CardGroup cols={2}>
2927 | <Card title="Inspector Repository" icon="github" href="https://github.com/modelcontextprotocol/inspector">
2928 | Check out the MCP Inspector source code
2929 | </Card>
2930 |
2931 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
2932 | Learn about broader debugging strategies
2933 | </Card>
2934 | </CardGroup>
2935 |
2936 |
2937 | # Example Servers
2938 | Source: https://modelcontextprotocol.io/examples
2939 |
2940 | A list of example servers and implementations
2941 |
2942 | 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.
2943 |
2944 | ## Reference implementations
2945 |
2946 | These official reference servers demonstrate core MCP features and SDK usage:
2947 |
2948 | ### Data and file systems
2949 |
2950 | * **[Filesystem](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem)** - Secure file operations with configurable access controls
2951 | * **[PostgreSQL](https://github.com/modelcontextprotocol/servers/tree/main/src/postgres)** - Read-only database access with schema inspection capabilities
2952 | * **[SQLite](https://github.com/modelcontextprotocol/servers/tree/main/src/sqlite)** - Database interaction and business intelligence features
2953 | * **[Google Drive](https://github.com/modelcontextprotocol/servers/tree/main/src/gdrive)** - File access and search capabilities for Google Drive
2954 |
2955 | ### Development tools
2956 |
2957 | * **[Git](https://github.com/modelcontextprotocol/servers/tree/main/src/git)** - Tools to read, search, and manipulate Git repositories
2958 | * **[GitHub](https://github.com/modelcontextprotocol/servers/tree/main/src/github)** - Repository management, file operations, and GitHub API integration
2959 | * **[GitLab](https://github.com/modelcontextprotocol/servers/tree/main/src/gitlab)** - GitLab API integration enabling project management
2960 | * **[Sentry](https://github.com/modelcontextprotocol/servers/tree/main/src/sentry)** - Retrieving and analyzing issues from Sentry.io
2961 |
2962 | ### Web and browser automation
2963 |
2964 | * **[Brave Search](https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search)** - Web and local search using Brave's Search API
2965 | * **[Fetch](https://github.com/modelcontextprotocol/servers/tree/main/src/fetch)** - Web content fetching and conversion optimized for LLM usage
2966 | * **[Puppeteer](https://github.com/modelcontextprotocol/servers/tree/main/src/puppeteer)** - Browser automation and web scraping capabilities
2967 |
2968 | ### Productivity and communication
2969 |
2970 | * **[Slack](https://github.com/modelcontextprotocol/servers/tree/main/src/slack)** - Channel management and messaging capabilities
2971 | * **[Google Maps](https://github.com/modelcontextprotocol/servers/tree/main/src/google-maps)** - Location services, directions, and place details
2972 | * **[Memory](https://github.com/modelcontextprotocol/servers/tree/main/src/memory)** - Knowledge graph-based persistent memory system
2973 |
2974 | ### AI and specialized tools
2975 |
2976 | * **[EverArt](https://github.com/modelcontextprotocol/servers/tree/main/src/everart)** - AI image generation using various models
2977 | * **[Sequential Thinking](https://github.com/modelcontextprotocol/servers/tree/main/src/sequentialthinking)** - Dynamic problem-solving through thought sequences
2978 | * **[AWS KB Retrieval](https://github.com/modelcontextprotocol/servers/tree/main/src/aws-kb-retrieval-server)** - Retrieval from AWS Knowledge Base using Bedrock Agent Runtime
2979 |
2980 | ## Official integrations
2981 |
2982 | These MCP servers are maintained by companies for their platforms:
2983 |
2984 | * **[Axiom](https://github.com/axiomhq/mcp-server-axiom)** - Query and analyze logs, traces, and event data using natural language
2985 | * **[Browserbase](https://github.com/browserbase/mcp-server-browserbase)** - Automate browser interactions in the cloud
2986 | * **[Cloudflare](https://github.com/cloudflare/mcp-server-cloudflare)** - Deploy and manage resources on the Cloudflare developer platform
2987 | * **[E2B](https://github.com/e2b-dev/mcp-server)** - Execute code in secure cloud sandboxes
2988 | * **[Neon](https://github.com/neondatabase/mcp-server-neon)** - Interact with the Neon serverless Postgres platform
2989 | * **[Obsidian Markdown Notes](https://github.com/calclavia/mcp-obsidian)** - Read and search through Markdown notes in Obsidian vaults
2990 | * **[Qdrant](https://github.com/qdrant/mcp-server-qdrant/)** - Implement semantic memory using the Qdrant vector search engine
2991 | * **[Raygun](https://github.com/MindscapeHQ/mcp-server-raygun)** - Access crash reporting and monitoring data
2992 | * **[Search1API](https://github.com/fatwang2/search1api-mcp)** - Unified API for search, crawling, and sitemaps
2993 | * **[Stripe](https://github.com/stripe/agent-toolkit)** - Interact with the Stripe API
2994 | * **[Tinybird](https://github.com/tinybirdco/mcp-tinybird)** - Interface with the Tinybird serverless ClickHouse platform
2995 |
2996 | ## Community highlights
2997 |
2998 | A growing ecosystem of community-developed servers extends MCP's capabilities:
2999 |
3000 | * **[Docker](https://github.com/ckreiling/mcp-server-docker)** - Manage containers, images, volumes, and networks
3001 | * **[Kubernetes](https://github.com/Flux159/mcp-server-kubernetes)** - Manage pods, deployments, and services
3002 | * **[Linear](https://github.com/jerhadf/linear-mcp-server)** - Project management and issue tracking
3003 | * **[Snowflake](https://github.com/datawiz168/mcp-snowflake-service)** - Interact with Snowflake databases
3004 | * **[Spotify](https://github.com/varunneal/spotify-mcp)** - Control Spotify playback and manage playlists
3005 | * **[Todoist](https://github.com/abhiz123/todoist-mcp-server)** - Task management integration
3006 |
3007 | > **Note:** Community servers are untested and should be used at your own risk. They are not affiliated with or endorsed by Anthropic.
3008 |
3009 | For a complete list of community servers, visit the [MCP Servers Repository](https://github.com/modelcontextprotocol/servers).
3010 |
3011 | ## Getting started
3012 |
3013 | ### Using reference servers
3014 |
3015 | TypeScript-based servers can be used directly with `npx`:
3016 |
3017 | ```bash
3018 | npx -y @modelcontextprotocol/server-memory
3019 | ```
3020 |
3021 | Python-based servers can be used with `uvx` (recommended) or `pip`:
3022 |
3023 | ```bash
3024 | # Using uvx
3025 | uvx mcp-server-git
3026 |
3027 | # Using pip
3028 | pip install mcp-server-git
3029 | python -m mcp_server_git
3030 | ```
3031 |
3032 | ### Configuring with Claude
3033 |
3034 | To use an MCP server with Claude, add it to your configuration:
3035 |
3036 | ```json
3037 | {
3038 | "mcpServers": {
3039 | "memory": {
3040 | "command": "npx",
3041 | "args": ["-y", "@modelcontextprotocol/server-memory"]
3042 | },
3043 | "filesystem": {
3044 | "command": "npx",
3045 | "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/allowed/files"]
3046 | },
3047 | "github": {
3048 | "command": "npx",
3049 | "args": ["-y", "@modelcontextprotocol/server-github"],
3050 | "env": {
3051 | "GITHUB_PERSONAL_ACCESS_TOKEN": "<YOUR_TOKEN>"
3052 | }
3053 | }
3054 | }
3055 | }
3056 | ```
3057 |
3058 | ## Additional resources
3059 |
3060 | * [MCP Servers Repository](https://github.com/modelcontextprotocol/servers) - Complete collection of reference implementations and community servers
3061 | * [Awesome MCP Servers](https://github.com/punkpeye/awesome-mcp-servers) - Curated list of MCP servers
3062 | * [MCP CLI](https://github.com/wong2/mcp-cli) - Command-line inspector for testing MCP servers
3063 | * [MCP Get](https://mcp-get.com) - Tool for installing and managing MCP servers
3064 | * [Supergateway](https://github.com/supercorp-ai/supergateway) - Run MCP stdio servers over SSE
3065 |
3066 | Visit our [GitHub Discussions](https://github.com/orgs/modelcontextprotocol/discussions) to engage with the MCP community.
3067 |
3068 |
3069 | # Introduction
3070 | Source: https://modelcontextprotocol.io/introduction
3071 |
3072 | Get started with the Model Context Protocol (MCP)
3073 |
3074 | <Note>Java SDK released! Check out [what else is new.](/development/updates)</Note>
3075 |
3076 | 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.
3077 |
3078 | ## Why MCP?
3079 |
3080 | MCP helps you build agents and complex workflows on top of LLMs. LLMs frequently need to integrate with data and tools, and MCP provides:
3081 |
3082 | * A growing list of pre-built integrations that your LLM can directly plug into
3083 | * The flexibility to switch between LLM providers and vendors
3084 | * Best practices for securing your data within your infrastructure
3085 |
3086 | ### General architecture
3087 |
3088 | At its core, MCP follows a client-server architecture where a host application can connect to multiple servers:
3089 |
3090 | ```mermaid
3091 | flowchart LR
3092 | subgraph "Your Computer"
3093 | Host["Host with MCP Client\n(Claude, IDEs, Tools)"]
3094 | S1["MCP Server A"]
3095 | S2["MCP Server B"]
3096 | S3["MCP Server C"]
3097 | Host <-->|"MCP Protocol"| S1
3098 | Host <-->|"MCP Protocol"| S2
3099 | Host <-->|"MCP Protocol"| S3
3100 | S1 <--> D1[("Local\nData Source A")]
3101 | S2 <--> D2[("Local\nData Source B")]
3102 | end
3103 | subgraph "Internet"
3104 | S3 <-->|"Web APIs"| D3[("Remote\nService C")]
3105 | end
3106 | ```
3107 |
3108 | * **MCP Hosts**: Programs like Claude Desktop, IDEs, or AI tools that want to access data through MCP
3109 | * **MCP Clients**: Protocol clients that maintain 1:1 connections with servers
3110 | * **MCP Servers**: Lightweight programs that each expose specific capabilities through the standardized Model Context Protocol
3111 | * **Local Data Sources**: Your computer's files, databases, and services that MCP servers can securely access
3112 | * **Remote Services**: External systems available over the internet (e.g., through APIs) that MCP servers can connect to
3113 |
3114 | ## Get started
3115 |
3116 | Choose the path that best fits your needs:
3117 |
3118 | #### Quick Starts
3119 |
3120 | <CardGroup cols={2}>
3121 | <Card title="For Server Developers" icon="bolt" href="/quickstart/server">
3122 | Get started building your own server to use in Claude for Desktop and other clients
3123 | </Card>
3124 |
3125 | <Card title="For Client Developers" icon="bolt" href="/quickstart/client">
3126 | Get started building your own client that can integrate with all MCP servers
3127 | </Card>
3128 |
3129 | <Card title="For Claude Desktop Users" icon="bolt" href="/quickstart/user">
3130 | Get started using pre-built servers in Claude for Desktop
3131 | </Card>
3132 | </CardGroup>
3133 |
3134 | #### Examples
3135 |
3136 | <CardGroup cols={2}>
3137 | <Card title="Example Servers" icon="grid" href="/examples">
3138 | Check out our gallery of official MCP servers and implementations
3139 | </Card>
3140 |
3141 | <Card title="Example Clients" icon="cubes" href="/clients">
3142 | View the list of clients that support MCP integrations
3143 | </Card>
3144 | </CardGroup>
3145 |
3146 | ## Tutorials
3147 |
3148 | <CardGroup cols={2}>
3149 | <Card title="Building MCP with LLMs" icon="comments" href="/tutorials/building-mcp-with-llms">
3150 | Learn how to use LLMs like Claude to speed up your MCP development
3151 | </Card>
3152 |
3153 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
3154 | Learn how to effectively debug MCP servers and integrations
3155 | </Card>
3156 |
3157 | <Card title="MCP Inspector" icon="magnifying-glass" href="/docs/tools/inspector">
3158 | Test and inspect your MCP servers with our interactive debugging tool
3159 | </Card>
3160 | </CardGroup>
3161 |
3162 | ## Explore MCP
3163 |
3164 | Dive deeper into MCP's core concepts and capabilities:
3165 |
3166 | <CardGroup cols={2}>
3167 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3168 | Understand how MCP connects clients, servers, and LLMs
3169 | </Card>
3170 |
3171 | <Card title="Resources" icon="database" href="/docs/concepts/resources">
3172 | Expose data and content from your servers to LLMs
3173 | </Card>
3174 |
3175 | <Card title="Prompts" icon="message" href="/docs/concepts/prompts">
3176 | Create reusable prompt templates and workflows
3177 | </Card>
3178 |
3179 | <Card title="Tools" icon="wrench" href="/docs/concepts/tools">
3180 | Enable LLMs to perform actions through your server
3181 | </Card>
3182 |
3183 | <Card title="Sampling" icon="robot" href="/docs/concepts/sampling">
3184 | Let your servers request completions from LLMs
3185 | </Card>
3186 |
3187 | <Card title="Transports" icon="network-wired" href="/docs/concepts/transports">
3188 | Learn about MCP's communication mechanism
3189 | </Card>
3190 | </CardGroup>
3191 |
3192 | ## Contributing
3193 |
3194 | Want to contribute? Check out our [Contributing Guide](/development/contributing) to learn how you can help improve MCP.
3195 |
3196 | ## Support and Feedback
3197 |
3198 | Here's how to get help or provide feedback:
3199 |
3200 | * 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)
3201 | * For discussions or Q\&A about the MCP specification, use the [specification discussions](https://github.com/modelcontextprotocol/specification/discussions)
3202 | * For discussions or Q\&A about other MCP open source components, use the [organization discussions](https://github.com/orgs/modelcontextprotocol/discussions)
3203 | * For bug reports, feature requests, and questions related to Claude.app and claude.ai's MCP integration, please email [[email protected]](mailto:[email protected])
3204 |
3205 |
3206 | # For Client Developers
3207 | Source: https://modelcontextprotocol.io/quickstart/client
3208 |
3209 | Get started building your own client that can integrate with all MCP servers.
3210 |
3211 | 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.
3212 |
3213 | <Tabs>
3214 | <Tab title="Python">
3215 | [You can find the complete code for this tutorial here.](https://github.com/modelcontextprotocol/quickstart-resources/tree/main/mcp-client)
3216 |
3217 | ## System Requirements
3218 |
3219 | Before starting, ensure your system meets these requirements:
3220 |
3221 | * Mac or Windows computer
3222 | * Latest Python version installed
3223 | * Latest version of `uv` installed
3224 |
3225 | ## Setting Up Your Environment
3226 |
3227 | First, create a new Python project with `uv`:
3228 |
3229 | ```bash
3230 | # Create project directory
3231 | uv init mcp-client
3232 | cd mcp-client
3233 |
3234 | # Create virtual environment
3235 | uv venv
3236 |
3237 | # Activate virtual environment
3238 | # On Windows:
3239 | .venv\Scripts\activate
3240 | # On Unix or MacOS:
3241 | source .venv/bin/activate
3242 |
3243 | # Install required packages
3244 | uv add mcp anthropic python-dotenv
3245 |
3246 | # Remove boilerplate files
3247 | rm hello.py
3248 |
3249 | # Create our main file
3250 | touch client.py
3251 | ```
3252 |
3253 | ## Setting Up Your API Key
3254 |
3255 | You'll need an Anthropic API key from the [Anthropic Console](https://console.anthropic.com/settings/keys).
3256 |
3257 | Create a `.env` file to store it:
3258 |
3259 | ```bash
3260 | # Create .env file
3261 | touch .env
3262 | ```
3263 |
3264 | Add your key to the `.env` file:
3265 |
3266 | ```bash
3267 | ANTHROPIC_API_KEY=<your key here>
3268 | ```
3269 |
3270 | Add `.env` to your `.gitignore`:
3271 |
3272 | ```bash
3273 | echo ".env" >> .gitignore
3274 | ```
3275 |
3276 | <Warning>
3277 | Make sure you keep your `ANTHROPIC_API_KEY` secure!
3278 | </Warning>
3279 |
3280 | ## Creating the Client
3281 |
3282 | ### Basic Client Structure
3283 |
3284 | First, let's set up our imports and create the basic client class:
3285 |
3286 | ```python
3287 | import asyncio
3288 | from typing import Optional
3289 | from contextlib import AsyncExitStack
3290 |
3291 | from mcp import ClientSession, StdioServerParameters
3292 | from mcp.client.stdio import stdio_client
3293 |
3294 | from anthropic import Anthropic
3295 | from dotenv import load_dotenv
3296 |
3297 | load_dotenv() # load environment variables from .env
3298 |
3299 | class MCPClient:
3300 | def __init__(self):
3301 | # Initialize session and client objects
3302 | self.session: Optional[ClientSession] = None
3303 | self.exit_stack = AsyncExitStack()
3304 | self.anthropic = Anthropic()
3305 | # methods will go here
3306 | ```
3307 |
3308 | ### Server Connection Management
3309 |
3310 | Next, we'll implement the method to connect to an MCP server:
3311 |
3312 | ```python
3313 | async def connect_to_server(self, server_script_path: str):
3314 | """Connect to an MCP server
3315 |
3316 | Args:
3317 | server_script_path: Path to the server script (.py or .js)
3318 | """
3319 | is_python = server_script_path.endswith('.py')
3320 | is_js = server_script_path.endswith('.js')
3321 | if not (is_python or is_js):
3322 | raise ValueError("Server script must be a .py or .js file")
3323 |
3324 | command = "python" if is_python else "node"
3325 | server_params = StdioServerParameters(
3326 | command=command,
3327 | args=[server_script_path],
3328 | env=None
3329 | )
3330 |
3331 | stdio_transport = await self.exit_stack.enter_async_context(stdio_client(server_params))
3332 | self.stdio, self.write = stdio_transport
3333 | self.session = await self.exit_stack.enter_async_context(ClientSession(self.stdio, self.write))
3334 |
3335 | await self.session.initialize()
3336 |
3337 | # List available tools
3338 | response = await self.session.list_tools()
3339 | tools = response.tools
3340 | print("\nConnected to server with tools:", [tool.name for tool in tools])
3341 | ```
3342 |
3343 | ### Query Processing Logic
3344 |
3345 | Now let's add the core functionality for processing queries and handling tool calls:
3346 |
3347 | ```python
3348 | async def process_query(self, query: str) -> str:
3349 | """Process a query using Claude and available tools"""
3350 | messages = [
3351 | {
3352 | "role": "user",
3353 | "content": query
3354 | }
3355 | ]
3356 |
3357 | response = await self.session.list_tools()
3358 | available_tools = [{
3359 | "name": tool.name,
3360 | "description": tool.description,
3361 | "input_schema": tool.inputSchema
3362 | } for tool in response.tools]
3363 |
3364 | # Initial Claude API call
3365 | response = self.anthropic.messages.create(
3366 | model="claude-3-5-sonnet-20241022",
3367 | max_tokens=1000,
3368 | messages=messages,
3369 | tools=available_tools
3370 | )
3371 |
3372 | # Process response and handle tool calls
3373 | tool_results = []
3374 | final_text = []
3375 |
3376 | assistant_message_content = []
3377 | for content in response.content:
3378 | if content.type == 'text':
3379 | final_text.append(content.text)
3380 | assistant_message_content.append(content)
3381 | elif content.type == 'tool_use':
3382 | tool_name = content.name
3383 | tool_args = content.input
3384 |
3385 | # Execute tool call
3386 | result = await self.session.call_tool(tool_name, tool_args)
3387 | tool_results.append({"call": tool_name, "result": result})
3388 | final_text.append(f"[Calling tool {tool_name} with args {tool_args}]")
3389 |
3390 | assistant_message_content.append(content)
3391 | messages.append({
3392 | "role": "assistant",
3393 | "content": assistant_message_content
3394 | })
3395 | messages.append({
3396 | "role": "user",
3397 | "content": [
3398 | {
3399 | "type": "tool_result",
3400 | "tool_use_id": content.id,
3401 | "content": result.content
3402 | }
3403 | ]
3404 | })
3405 |
3406 | # Get next response from Claude
3407 | response = self.anthropic.messages.create(
3408 | model="claude-3-5-sonnet-20241022",
3409 | max_tokens=1000,
3410 | messages=messages,
3411 | tools=available_tools
3412 | )
3413 |
3414 | final_text.append(response.content[0].text)
3415 |
3416 | return "\n".join(final_text)
3417 | ```
3418 |
3419 | ### Interactive Chat Interface
3420 |
3421 | Now we'll add the chat loop and cleanup functionality:
3422 |
3423 | ```python
3424 | async def chat_loop(self):
3425 | """Run an interactive chat loop"""
3426 | print("\nMCP Client Started!")
3427 | print("Type your queries or 'quit' to exit.")
3428 |
3429 | while True:
3430 | try:
3431 | query = input("\nQuery: ").strip()
3432 |
3433 | if query.lower() == 'quit':
3434 | break
3435 |
3436 | response = await self.process_query(query)
3437 | print("\n" + response)
3438 |
3439 | except Exception as e:
3440 | print(f"\nError: {str(e)}")
3441 |
3442 | async def cleanup(self):
3443 | """Clean up resources"""
3444 | await self.exit_stack.aclose()
3445 | ```
3446 |
3447 | ### Main Entry Point
3448 |
3449 | Finally, we'll add the main execution logic:
3450 |
3451 | ```python
3452 | async def main():
3453 | if len(sys.argv) < 2:
3454 | print("Usage: python client.py <path_to_server_script>")
3455 | sys.exit(1)
3456 |
3457 | client = MCPClient()
3458 | try:
3459 | await client.connect_to_server(sys.argv[1])
3460 | await client.chat_loop()
3461 | finally:
3462 | await client.cleanup()
3463 |
3464 | if __name__ == "__main__":
3465 | import sys
3466 | asyncio.run(main())
3467 | ```
3468 |
3469 | You can find the complete `client.py` file [here.](https://gist.github.com/zckly/f3f28ea731e096e53b39b47bf0a2d4b1)
3470 |
3471 | ## Key Components Explained
3472 |
3473 | ### 1. Client Initialization
3474 |
3475 | * The `MCPClient` class initializes with session management and API clients
3476 | * Uses `AsyncExitStack` for proper resource management
3477 | * Configures the Anthropic client for Claude interactions
3478 |
3479 | ### 2. Server Connection
3480 |
3481 | * Supports both Python and Node.js servers
3482 | * Validates server script type
3483 | * Sets up proper communication channels
3484 | * Initializes the session and lists available tools
3485 |
3486 | ### 3. Query Processing
3487 |
3488 | * Maintains conversation context
3489 | * Handles Claude's responses and tool calls
3490 | * Manages the message flow between Claude and tools
3491 | * Combines results into a coherent response
3492 |
3493 | ### 4. Interactive Interface
3494 |
3495 | * Provides a simple command-line interface
3496 | * Handles user input and displays responses
3497 | * Includes basic error handling
3498 | * Allows graceful exit
3499 |
3500 | ### 5. Resource Management
3501 |
3502 | * Proper cleanup of resources
3503 | * Error handling for connection issues
3504 | * Graceful shutdown procedures
3505 |
3506 | ## Common Customization Points
3507 |
3508 | 1. **Tool Handling**
3509 | * Modify `process_query()` to handle specific tool types
3510 | * Add custom error handling for tool calls
3511 | * Implement tool-specific response formatting
3512 |
3513 | 2. **Response Processing**
3514 | * Customize how tool results are formatted
3515 | * Add response filtering or transformation
3516 | * Implement custom logging
3517 |
3518 | 3. **User Interface**
3519 | * Add a GUI or web interface
3520 | * Implement rich console output
3521 | * Add command history or auto-completion
3522 |
3523 | ## Running the Client
3524 |
3525 | To run your client with any MCP server:
3526 |
3527 | ```bash
3528 | uv run client.py path/to/server.py # python server
3529 | uv run client.py path/to/build/index.js # node server
3530 | ```
3531 |
3532 | <Note>
3533 | 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`
3534 | </Note>
3535 |
3536 | The client will:
3537 |
3538 | 1. Connect to the specified server
3539 | 2. List available tools
3540 | 3. Start an interactive chat session where you can:
3541 | * Enter queries
3542 | * See tool executions
3543 | * Get responses from Claude
3544 |
3545 | Here's an example of what it should look like if connected to the weather server from the server quickstart:
3546 |
3547 | <Frame>
3548 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/client-claude-cli-python.png" />
3549 | </Frame>
3550 |
3551 | ## How It Works
3552 |
3553 | When you submit a query:
3554 |
3555 | 1. The client gets the list of available tools from the server
3556 | 2. Your query is sent to Claude along with tool descriptions
3557 | 3. Claude decides which tools (if any) to use
3558 | 4. The client executes any requested tool calls through the server
3559 | 5. Results are sent back to Claude
3560 | 6. Claude provides a natural language response
3561 | 7. The response is displayed to you
3562 |
3563 | ## Best practices
3564 |
3565 | 1. **Error Handling**
3566 | * Always wrap tool calls in try-catch blocks
3567 | * Provide meaningful error messages
3568 | * Gracefully handle connection issues
3569 |
3570 | 2. **Resource Management**
3571 | * Use `AsyncExitStack` for proper cleanup
3572 | * Close connections when done
3573 | * Handle server disconnections
3574 |
3575 | 3. **Security**
3576 | * Store API keys securely in `.env`
3577 | * Validate server responses
3578 | * Be cautious with tool permissions
3579 |
3580 | ## Troubleshooting
3581 |
3582 | ### Server Path Issues
3583 |
3584 | * Double-check the path to your server script is correct
3585 | * Use the absolute path if the relative path isn't working
3586 | * For Windows users, make sure to use forward slashes (/) or escaped backslashes (\\) in the path
3587 | * Verify the server file has the correct extension (.py for Python or .js for Node.js)
3588 |
3589 | Example of correct path usage:
3590 |
3591 | ```bash
3592 | # Relative path
3593 | uv run client.py ./server/weather.py
3594 |
3595 | # Absolute path
3596 | uv run client.py /Users/username/projects/mcp-server/weather.py
3597 |
3598 | # Windows path (either format works)
3599 | uv run client.py C:/projects/mcp-server/weather.py
3600 | uv run client.py C:\\projects\\mcp-server\\weather.py
3601 | ```
3602 |
3603 | ### Response Timing
3604 |
3605 | * The first response might take up to 30 seconds to return
3606 | * This is normal and happens while:
3607 | * The server initializes
3608 | * Claude processes the query
3609 | * Tools are being executed
3610 | * Subsequent responses are typically faster
3611 | * Don't interrupt the process during this initial waiting period
3612 |
3613 | ### Common Error Messages
3614 |
3615 | If you see:
3616 |
3617 | * `FileNotFoundError`: Check your server path
3618 | * `Connection refused`: Ensure the server is running and the path is correct
3619 | * `Tool execution failed`: Verify the tool's required environment variables are set
3620 | * `Timeout error`: Consider increasing the timeout in your client configuration
3621 | </Tab>
3622 |
3623 | <Tab title="Java">
3624 | <Note>
3625 | This is a quickstart demo based on Spring AI MCP auto-configuration and boot starters.
3626 | To learn how to create sync and async MCP Clients manually, consult the [Java SDK Client](/sdk/java/mcp-client) documentation
3627 | </Note>
3628 |
3629 | 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.
3630 | [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)
3631 |
3632 | ## System Requirements
3633 |
3634 | Before starting, ensure your system meets these requirements:
3635 |
3636 | * Java 17 or higher
3637 | * Maven 3.6+
3638 | * npx package manager
3639 | * Anthropic API key (Claude)
3640 | * Brave Search API key
3641 |
3642 | ## Setting Up Your Environment
3643 |
3644 | 1. Install npx (Node Package eXecute):
3645 | First, make sure to install [npm](https://docs.npmjs.com/downloading-and-installing-node-js-and-npm)
3646 | and then run:
3647 | ```bash
3648 | npm install -g npx
3649 | ```
3650 |
3651 | 2. Clone the repository:
3652 | ```bash
3653 | git clone https://github.com/spring-projects/spring-ai-examples.git
3654 | cd model-context-protocol/brave-chatbot
3655 | ```
3656 |
3657 | 3. Set up your API keys:
3658 | ```bash
3659 | export ANTHROPIC_API_KEY='your-anthropic-api-key-here'
3660 | export BRAVE_API_KEY='your-brave-api-key-here'
3661 | ```
3662 |
3663 | 4. Build the application:
3664 | ```bash
3665 | ./mvnw clean install
3666 | ```
3667 |
3668 | 5. Run the application using Maven:
3669 | ```bash
3670 | ./mvnw spring-boot:run
3671 | ```
3672 |
3673 | <Warning>
3674 | Make sure you keep your `ANTHROPIC_API_KEY` and `BRAVE_API_KEY` keys secure!
3675 | </Warning>
3676 |
3677 | ## How it Works
3678 |
3679 | The application integrates Spring AI with the Brave Search MCP server through several components:
3680 |
3681 | ### MCP Client Configuration
3682 |
3683 | 1. Required dependencies in pom.xml:
3684 |
3685 | ```xml
3686 | <dependency>
3687 | <groupId>org.springframework.ai</groupId>
3688 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
3689 | </dependency>
3690 | <dependency>
3691 | <groupId>org.springframework.ai</groupId>
3692 | <artifactId>spring-ai-anthropic-spring-boot-starter</artifactId>
3693 | </dependency>
3694 | ```
3695 |
3696 | 2. Application properties (application.yml):
3697 |
3698 | ```yml
3699 | spring:
3700 | ai:
3701 | mcp:
3702 | client:
3703 | enabled: true
3704 | name: brave-search-client
3705 | version: 1.0.0
3706 | type: SYNC
3707 | request-timeout: 20s
3708 | stdio:
3709 | root-change-notification: true
3710 | servers-configuration: classpath:/mcp-servers-config.json
3711 | anthropic:
3712 | api-key: ${ANTHROPIC_API_KEY}
3713 | ```
3714 |
3715 | This activates the `spring-ai-mcp-client-spring-boot-starter` to create one or more `McpClient`s based on the provided server configuration.
3716 |
3717 | 3. MCP Server Configuration (`mcp-servers-config.json`):
3718 |
3719 | ```json
3720 | {
3721 | "mcpServers": {
3722 | "brave-search": {
3723 | "command": "npx",
3724 | "args": [
3725 | "-y",
3726 | "@modelcontextprotocol/server-brave-search"
3727 | ],
3728 | "env": {
3729 | "BRAVE_API_KEY": "<PUT YOUR BRAVE API KEY>"
3730 | }
3731 | }
3732 | }
3733 | }
3734 | ```
3735 |
3736 | ### Chat Implementation
3737 |
3738 | The chatbot is implemented using Spring AI's ChatClient with MCP tool integration:
3739 |
3740 | ```java
3741 | var chatClient = chatClientBuilder
3742 | .defaultSystem("You are useful assistant, expert in AI and Java.")
3743 | .defaultTools((Object[]) mcpToolAdapter.toolCallbacks())
3744 | .defaultAdvisors(new MessageChatMemoryAdvisor(new InMemoryChatMemory()))
3745 | .build();
3746 | ```
3747 |
3748 | Key features:
3749 |
3750 | * Uses Claude AI model for natural language understanding
3751 | * Integrates Brave Search through MCP for real-time web search capabilities
3752 | * Maintains conversation memory using InMemoryChatMemory
3753 | * Runs as an interactive command-line application
3754 |
3755 | ### Build and run
3756 |
3757 | ```bash
3758 | ./mvnw clean install
3759 | java -jar ./target/ai-mcp-brave-chatbot-0.0.1-SNAPSHOT.jar
3760 | ```
3761 |
3762 | or
3763 |
3764 | ```bash
3765 | ./mvnw spring-boot:run
3766 | ```
3767 |
3768 | 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.
3769 |
3770 | The chatbot can:
3771 |
3772 | * Answer questions using its built-in knowledge
3773 | * Perform web searches when needed using Brave Search
3774 | * Remember context from previous messages in the conversation
3775 | * Combine information from multiple sources to provide comprehensive answers
3776 |
3777 | ### Advanced Configuration
3778 |
3779 | The MCP client supports additional configuration options:
3780 |
3781 | * Client customization through `McpSyncClientCustomizer` or `McpAsyncClientCustomizer`
3782 | * Multiple clients with multiple transport types: `STDIO` and `SSE` (Server-Sent Events)
3783 | * Integration with Spring AI's tool execution framework
3784 | * Automatic client initialization and lifecycle management
3785 |
3786 | For WebFlux-based applications, you can use the WebFlux starter instead:
3787 |
3788 | ```xml
3789 | <dependency>
3790 | <groupId>org.springframework.ai</groupId>
3791 | <artifactId>spring-ai-mcp-client-webflux-spring-boot-starter</artifactId>
3792 | </dependency>
3793 | ```
3794 |
3795 | This provides similar functionality but uses a WebFlux-based SSE transport implementation, recommended for production deployments.
3796 | </Tab>
3797 | </Tabs>
3798 |
3799 | ## Next steps
3800 |
3801 | <CardGroup cols={2}>
3802 | <Card title="Example servers" icon="grid" href="/examples">
3803 | Check out our gallery of official MCP servers and implementations
3804 | </Card>
3805 |
3806 | <Card title="Clients" icon="cubes" href="/clients">
3807 | View the list of clients that support MCP integrations
3808 | </Card>
3809 |
3810 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
3811 | Learn how to use LLMs like Claude to speed up your MCP development
3812 | </Card>
3813 |
3814 | <Card title="Core architecture" icon="sitemap" href="/docs/concepts/architecture">
3815 | Understand how MCP connects clients, servers, and LLMs
3816 | </Card>
3817 | </CardGroup>
3818 |
3819 |
3820 | # For Server Developers
3821 | Source: https://modelcontextprotocol.io/quickstart/server
3822 |
3823 | Get started building your own server to use in Claude for Desktop and other clients.
3824 |
3825 | 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.
3826 |
3827 | ### What we'll be building
3828 |
3829 | 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!
3830 |
3831 | 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):
3832 |
3833 | <Frame>
3834 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
3835 | </Frame>
3836 |
3837 | <Frame>
3838 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
3839 | </Frame>
3840 |
3841 | <Note>
3842 | 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).
3843 | </Note>
3844 |
3845 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
3846 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
3847 | </Accordion>
3848 |
3849 | ### Core MCP Concepts
3850 |
3851 | MCP servers can provide three main types of capabilities:
3852 |
3853 | 1. **Resources**: File-like data that can be read by clients (like API responses or file contents)
3854 | 2. **Tools**: Functions that can be called by the LLM (with user approval)
3855 | 3. **Prompts**: Pre-written templates that help users accomplish specific tasks
3856 |
3857 | This tutorial will primarily focus on tools.
3858 |
3859 | <Tabs>
3860 | <Tab title="Python">
3861 | 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)
3862 |
3863 | ### Prerequisite knowledge
3864 |
3865 | This quickstart assumes you have familiarity with:
3866 |
3867 | * Python
3868 | * LLMs like Claude
3869 |
3870 | ### System requirements
3871 |
3872 | * Python 3.10 or higher installed.
3873 | * You must use the Python MCP SDK 1.2.0 or higher.
3874 |
3875 | ### Set up your environment
3876 |
3877 | First, let's install `uv` and set up our Python project and environment:
3878 |
3879 | <CodeGroup>
3880 | ```bash MacOS/Linux
3881 | curl -LsSf https://astral.sh/uv/install.sh | sh
3882 | ```
3883 |
3884 | ```powershell Windows
3885 | powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
3886 | ```
3887 | </CodeGroup>
3888 |
3889 | Make sure to restart your terminal afterwards to ensure that the `uv` command gets picked up.
3890 |
3891 | Now, let's create and set up our project:
3892 |
3893 | <CodeGroup>
3894 | ```bash MacOS/Linux
3895 | # Create a new directory for our project
3896 | uv init weather
3897 | cd weather
3898 |
3899 | # Create virtual environment and activate it
3900 | uv venv
3901 | source .venv/bin/activate
3902 |
3903 | # Install dependencies
3904 | uv add "mcp[cli]" httpx
3905 |
3906 | # Create our server file
3907 | touch weather.py
3908 | ```
3909 |
3910 | ```powershell Windows
3911 | # Create a new directory for our project
3912 | uv init weather
3913 | cd weather
3914 |
3915 | # Create virtual environment and activate it
3916 | uv venv
3917 | .venv\Scripts\activate
3918 |
3919 | # Install dependencies
3920 | uv add mcp[cli] httpx
3921 |
3922 | # Create our server file
3923 | new-item weather.py
3924 | ```
3925 | </CodeGroup>
3926 |
3927 | Now let's dive into building your server.
3928 |
3929 | ## Building your server
3930 |
3931 | ### Importing packages and setting up the instance
3932 |
3933 | Add these to the top of your `weather.py`:
3934 |
3935 | ```python
3936 | from typing import Any
3937 | import httpx
3938 | from mcp.server.fastmcp import FastMCP
3939 |
3940 | # Initialize FastMCP server
3941 | mcp = FastMCP("weather")
3942 |
3943 | # Constants
3944 | NWS_API_BASE = "https://api.weather.gov"
3945 | USER_AGENT = "weather-app/1.0"
3946 | ```
3947 |
3948 | The FastMCP class uses Python type hints and docstrings to automatically generate tool definitions, making it easy to create and maintain MCP tools.
3949 |
3950 | ### Helper functions
3951 |
3952 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
3953 |
3954 | ```python
3955 | async def make_nws_request(url: str) -> dict[str, Any] | None:
3956 | """Make a request to the NWS API with proper error handling."""
3957 | headers = {
3958 | "User-Agent": USER_AGENT,
3959 | "Accept": "application/geo+json"
3960 | }
3961 | async with httpx.AsyncClient() as client:
3962 | try:
3963 | response = await client.get(url, headers=headers, timeout=30.0)
3964 | response.raise_for_status()
3965 | return response.json()
3966 | except Exception:
3967 | return None
3968 |
3969 | def format_alert(feature: dict) -> str:
3970 | """Format an alert feature into a readable string."""
3971 | props = feature["properties"]
3972 | return f"""
3973 | Event: {props.get('event', 'Unknown')}
3974 | Area: {props.get('areaDesc', 'Unknown')}
3975 | Severity: {props.get('severity', 'Unknown')}
3976 | Description: {props.get('description', 'No description available')}
3977 | Instructions: {props.get('instruction', 'No specific instructions provided')}
3978 | """
3979 | ```
3980 |
3981 | ### Implementing tool execution
3982 |
3983 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
3984 |
3985 | ```python
3986 | @mcp.tool()
3987 | async def get_alerts(state: str) -> str:
3988 | """Get weather alerts for a US state.
3989 |
3990 | Args:
3991 | state: Two-letter US state code (e.g. CA, NY)
3992 | """
3993 | url = f"{NWS_API_BASE}/alerts/active/area/{state}"
3994 | data = await make_nws_request(url)
3995 |
3996 | if not data or "features" not in data:
3997 | return "Unable to fetch alerts or no alerts found."
3998 |
3999 | if not data["features"]:
4000 | return "No active alerts for this state."
4001 |
4002 | alerts = [format_alert(feature) for feature in data["features"]]
4003 | return "\n---\n".join(alerts)
4004 |
4005 | @mcp.tool()
4006 | async def get_forecast(latitude: float, longitude: float) -> str:
4007 | """Get weather forecast for a location.
4008 |
4009 | Args:
4010 | latitude: Latitude of the location
4011 | longitude: Longitude of the location
4012 | """
4013 | # First get the forecast grid endpoint
4014 | points_url = f"{NWS_API_BASE}/points/{latitude},{longitude}"
4015 | points_data = await make_nws_request(points_url)
4016 |
4017 | if not points_data:
4018 | return "Unable to fetch forecast data for this location."
4019 |
4020 | # Get the forecast URL from the points response
4021 | forecast_url = points_data["properties"]["forecast"]
4022 | forecast_data = await make_nws_request(forecast_url)
4023 |
4024 | if not forecast_data:
4025 | return "Unable to fetch detailed forecast."
4026 |
4027 | # Format the periods into a readable forecast
4028 | periods = forecast_data["properties"]["periods"]
4029 | forecasts = []
4030 | for period in periods[:5]: # Only show next 5 periods
4031 | forecast = f"""
4032 | {period['name']}:
4033 | Temperature: {period['temperature']}°{period['temperatureUnit']}
4034 | Wind: {period['windSpeed']} {period['windDirection']}
4035 | Forecast: {period['detailedForecast']}
4036 | """
4037 | forecasts.append(forecast)
4038 |
4039 | return "\n---\n".join(forecasts)
4040 | ```
4041 |
4042 | ### Running the server
4043 |
4044 | Finally, let's initialize and run the server:
4045 |
4046 | ```python
4047 | if __name__ == "__main__":
4048 | # Initialize and run the server
4049 | mcp.run(transport='stdio')
4050 | ```
4051 |
4052 | Your server is complete! Run `uv run weather.py` to confirm that everything's working.
4053 |
4054 | Let's now test your server from an existing MCP host, Claude for Desktop.
4055 |
4056 | ## Testing your server with Claude for Desktop
4057 |
4058 | <Note>
4059 | 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.
4060 | </Note>
4061 |
4062 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4063 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4064 |
4065 | 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.
4066 |
4067 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4068 |
4069 | <Tabs>
4070 | <Tab title="MacOS/Linux">
4071 | ```bash
4072 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4073 | ```
4074 | </Tab>
4075 |
4076 | <Tab title="Windows">
4077 | ```powershell
4078 | code $env:AppData\Claude\claude_desktop_config.json
4079 | ```
4080 | </Tab>
4081 | </Tabs>
4082 |
4083 | 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.
4084 |
4085 | In this case, we'll add our single weather server like so:
4086 |
4087 | <Tabs>
4088 | <Tab title="MacOS/Linux">
4089 | ```json Python
4090 | {
4091 | "mcpServers": {
4092 | "weather": {
4093 | "command": "uv",
4094 | "args": [
4095 | "--directory",
4096 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather",
4097 | "run",
4098 | "weather.py"
4099 | ]
4100 | }
4101 | }
4102 | }
4103 | ```
4104 | </Tab>
4105 |
4106 | <Tab title="Windows">
4107 | ```json Python
4108 | {
4109 | "mcpServers": {
4110 | "weather": {
4111 | "command": "uv",
4112 | "args": [
4113 | "--directory",
4114 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather",
4115 | "run",
4116 | "weather.py"
4117 | ]
4118 | }
4119 | }
4120 | }
4121 | ```
4122 | </Tab>
4123 | </Tabs>
4124 |
4125 | <Warning>
4126 | 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.
4127 | </Warning>
4128 |
4129 | <Note>
4130 | Make sure you pass in the absolute path to your server.
4131 | </Note>
4132 |
4133 | This tells Claude for Desktop:
4134 |
4135 | 1. There's an MCP server named "weather"
4136 | 2. To launch it by running `uv --directory /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather run weather`
4137 |
4138 | Save the file, and restart **Claude for Desktop**.
4139 | </Tab>
4140 |
4141 | <Tab title="Node">
4142 | 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)
4143 |
4144 | ### Prerequisite knowledge
4145 |
4146 | This quickstart assumes you have familiarity with:
4147 |
4148 | * TypeScript
4149 | * LLMs like Claude
4150 |
4151 | ### System requirements
4152 |
4153 | For TypeScript, make sure you have the latest version of Node installed.
4154 |
4155 | ### Set up your environment
4156 |
4157 | First, let's install Node.js and npm if you haven't already. You can download them from [nodejs.org](https://nodejs.org/).
4158 | Verify your Node.js installation:
4159 |
4160 | ```bash
4161 | node --version
4162 | npm --version
4163 | ```
4164 |
4165 | For this tutorial, you'll need Node.js version 16 or higher.
4166 |
4167 | Now, let's create and set up our project:
4168 |
4169 | <CodeGroup>
4170 | ```bash MacOS/Linux
4171 | # Create a new directory for our project
4172 | mkdir weather
4173 | cd weather
4174 |
4175 | # Initialize a new npm project
4176 | npm init -y
4177 |
4178 | # Install dependencies
4179 | npm install @modelcontextprotocol/sdk zod
4180 | npm install -D @types/node typescript
4181 |
4182 | # Create our files
4183 | mkdir src
4184 | touch src/index.ts
4185 | ```
4186 |
4187 | ```powershell Windows
4188 | # Create a new directory for our project
4189 | md weather
4190 | cd weather
4191 |
4192 | # Initialize a new npm project
4193 | npm init -y
4194 |
4195 | # Install dependencies
4196 | npm install @modelcontextprotocol/sdk zod
4197 | npm install -D @types/node typescript
4198 |
4199 | # Create our files
4200 | md src
4201 | new-item src\index.ts
4202 | ```
4203 | </CodeGroup>
4204 |
4205 | Update your package.json to add type: "module" and a build script:
4206 |
4207 | ```json package.json
4208 | {
4209 | "type": "module",
4210 | "bin": {
4211 | "weather": "./build/index.js"
4212 | },
4213 | "scripts": {
4214 | "build": "tsc && node -e \"require('fs').chmodSync('build/index.js', '755')\"",
4215 | },
4216 | "files": [
4217 | "build"
4218 | ],
4219 | }
4220 | ```
4221 |
4222 | Create a `tsconfig.json` in the root of your project:
4223 |
4224 | ```json tsconfig.json
4225 | {
4226 | "compilerOptions": {
4227 | "target": "ES2022",
4228 | "module": "Node16",
4229 | "moduleResolution": "Node16",
4230 | "outDir": "./build",
4231 | "rootDir": "./src",
4232 | "strict": true,
4233 | "esModuleInterop": true,
4234 | "skipLibCheck": true,
4235 | "forceConsistentCasingInFileNames": true
4236 | },
4237 | "include": ["src/**/*"],
4238 | "exclude": ["node_modules"]
4239 | }
4240 | ```
4241 |
4242 | Now let's dive into building your server.
4243 |
4244 | ## Building your server
4245 |
4246 | ### Importing packages and setting up the instance
4247 |
4248 | Add these to the top of your `src/index.ts`:
4249 |
4250 | ```typescript
4251 | import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
4252 | import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
4253 | import { z } from "zod";
4254 |
4255 | const NWS_API_BASE = "https://api.weather.gov";
4256 | const USER_AGENT = "weather-app/1.0";
4257 |
4258 | // Create server instance
4259 | const server = new McpServer({
4260 | name: "weather",
4261 | version: "1.0.0",
4262 | });
4263 | ```
4264 |
4265 | ### Helper functions
4266 |
4267 | Next, let's add our helper functions for querying and formatting the data from the National Weather Service API:
4268 |
4269 | ```typescript
4270 | // Helper function for making NWS API requests
4271 | async function makeNWSRequest<T>(url: string): Promise<T | null> {
4272 | const headers = {
4273 | "User-Agent": USER_AGENT,
4274 | Accept: "application/geo+json",
4275 | };
4276 |
4277 | try {
4278 | const response = await fetch(url, { headers });
4279 | if (!response.ok) {
4280 | throw new Error(`HTTP error! status: ${response.status}`);
4281 | }
4282 | return (await response.json()) as T;
4283 | } catch (error) {
4284 | console.error("Error making NWS request:", error);
4285 | return null;
4286 | }
4287 | }
4288 |
4289 | interface AlertFeature {
4290 | properties: {
4291 | event?: string;
4292 | areaDesc?: string;
4293 | severity?: string;
4294 | status?: string;
4295 | headline?: string;
4296 | };
4297 | }
4298 |
4299 | // Format alert data
4300 | function formatAlert(feature: AlertFeature): string {
4301 | const props = feature.properties;
4302 | return [
4303 | `Event: ${props.event || "Unknown"}`,
4304 | `Area: ${props.areaDesc || "Unknown"}`,
4305 | `Severity: ${props.severity || "Unknown"}`,
4306 | `Status: ${props.status || "Unknown"}`,
4307 | `Headline: ${props.headline || "No headline"}`,
4308 | "---",
4309 | ].join("\n");
4310 | }
4311 |
4312 | interface ForecastPeriod {
4313 | name?: string;
4314 | temperature?: number;
4315 | temperatureUnit?: string;
4316 | windSpeed?: string;
4317 | windDirection?: string;
4318 | shortForecast?: string;
4319 | }
4320 |
4321 | interface AlertsResponse {
4322 | features: AlertFeature[];
4323 | }
4324 |
4325 | interface PointsResponse {
4326 | properties: {
4327 | forecast?: string;
4328 | };
4329 | }
4330 |
4331 | interface ForecastResponse {
4332 | properties: {
4333 | periods: ForecastPeriod[];
4334 | };
4335 | }
4336 | ```
4337 |
4338 | ### Implementing tool execution
4339 |
4340 | The tool execution handler is responsible for actually executing the logic of each tool. Let's add it:
4341 |
4342 | ```typescript
4343 | // Register weather tools
4344 | server.tool(
4345 | "get-alerts",
4346 | "Get weather alerts for a state",
4347 | {
4348 | state: z.string().length(2).describe("Two-letter state code (e.g. CA, NY)"),
4349 | },
4350 | async ({ state }) => {
4351 | const stateCode = state.toUpperCase();
4352 | const alertsUrl = `${NWS_API_BASE}/alerts?area=${stateCode}`;
4353 | const alertsData = await makeNWSRequest<AlertsResponse>(alertsUrl);
4354 |
4355 | if (!alertsData) {
4356 | return {
4357 | content: [
4358 | {
4359 | type: "text",
4360 | text: "Failed to retrieve alerts data",
4361 | },
4362 | ],
4363 | };
4364 | }
4365 |
4366 | const features = alertsData.features || [];
4367 | if (features.length === 0) {
4368 | return {
4369 | content: [
4370 | {
4371 | type: "text",
4372 | text: `No active alerts for ${stateCode}`,
4373 | },
4374 | ],
4375 | };
4376 | }
4377 |
4378 | const formattedAlerts = features.map(formatAlert);
4379 | const alertsText = `Active alerts for ${stateCode}:\n\n${formattedAlerts.join("\n")}`;
4380 |
4381 | return {
4382 | content: [
4383 | {
4384 | type: "text",
4385 | text: alertsText,
4386 | },
4387 | ],
4388 | };
4389 | },
4390 | );
4391 |
4392 | server.tool(
4393 | "get-forecast",
4394 | "Get weather forecast for a location",
4395 | {
4396 | latitude: z.number().min(-90).max(90).describe("Latitude of the location"),
4397 | longitude: z.number().min(-180).max(180).describe("Longitude of the location"),
4398 | },
4399 | async ({ latitude, longitude }) => {
4400 | // Get grid point data
4401 | const pointsUrl = `${NWS_API_BASE}/points/${latitude.toFixed(4)},${longitude.toFixed(4)}`;
4402 | const pointsData = await makeNWSRequest<PointsResponse>(pointsUrl);
4403 |
4404 | if (!pointsData) {
4405 | return {
4406 | content: [
4407 | {
4408 | type: "text",
4409 | text: `Failed to retrieve grid point data for coordinates: ${latitude}, ${longitude}. This location may not be supported by the NWS API (only US locations are supported).`,
4410 | },
4411 | ],
4412 | };
4413 | }
4414 |
4415 | const forecastUrl = pointsData.properties?.forecast;
4416 | if (!forecastUrl) {
4417 | return {
4418 | content: [
4419 | {
4420 | type: "text",
4421 | text: "Failed to get forecast URL from grid point data",
4422 | },
4423 | ],
4424 | };
4425 | }
4426 |
4427 | // Get forecast data
4428 | const forecastData = await makeNWSRequest<ForecastResponse>(forecastUrl);
4429 | if (!forecastData) {
4430 | return {
4431 | content: [
4432 | {
4433 | type: "text",
4434 | text: "Failed to retrieve forecast data",
4435 | },
4436 | ],
4437 | };
4438 | }
4439 |
4440 | const periods = forecastData.properties?.periods || [];
4441 | if (periods.length === 0) {
4442 | return {
4443 | content: [
4444 | {
4445 | type: "text",
4446 | text: "No forecast periods available",
4447 | },
4448 | ],
4449 | };
4450 | }
4451 |
4452 | // Format forecast periods
4453 | const formattedForecast = periods.map((period: ForecastPeriod) =>
4454 | [
4455 | `${period.name || "Unknown"}:`,
4456 | `Temperature: ${period.temperature || "Unknown"}°${period.temperatureUnit || "F"}`,
4457 | `Wind: ${period.windSpeed || "Unknown"} ${period.windDirection || ""}`,
4458 | `${period.shortForecast || "No forecast available"}`,
4459 | "---",
4460 | ].join("\n"),
4461 | );
4462 |
4463 | const forecastText = `Forecast for ${latitude}, ${longitude}:\n\n${formattedForecast.join("\n")}`;
4464 |
4465 | return {
4466 | content: [
4467 | {
4468 | type: "text",
4469 | text: forecastText,
4470 | },
4471 | ],
4472 | };
4473 | },
4474 | );
4475 | ```
4476 |
4477 | ### Running the server
4478 |
4479 | Finally, implement the main function to run the server:
4480 |
4481 | ```typescript
4482 | async function main() {
4483 | const transport = new StdioServerTransport();
4484 | await server.connect(transport);
4485 | console.error("Weather MCP Server running on stdio");
4486 | }
4487 |
4488 | main().catch((error) => {
4489 | console.error("Fatal error in main():", error);
4490 | process.exit(1);
4491 | });
4492 | ```
4493 |
4494 | Make sure to run `npm run build` to build your server! This is a very important step in getting your server to connect.
4495 |
4496 | Let's now test your server from an existing MCP host, Claude for Desktop.
4497 |
4498 | ## Testing your server with Claude for Desktop
4499 |
4500 | <Note>
4501 | 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.
4502 | </Note>
4503 |
4504 | First, make sure you have Claude for Desktop installed. [You can install the latest version
4505 | here.](https://claude.ai/download) If you already have Claude for Desktop, **make sure it's updated to the latest version.**
4506 |
4507 | 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.
4508 |
4509 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4510 |
4511 | <Tabs>
4512 | <Tab title="MacOS/Linux">
4513 | ```bash
4514 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4515 | ```
4516 | </Tab>
4517 |
4518 | <Tab title="Windows">
4519 | ```powershell
4520 | code $env:AppData\Claude\claude_desktop_config.json
4521 | ```
4522 | </Tab>
4523 | </Tabs>
4524 |
4525 | 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.
4526 |
4527 | In this case, we'll add our single weather server like so:
4528 |
4529 | <Tabs>
4530 | <Tab title="MacOS/Linux">
4531 | <CodeGroup>
4532 | ```json Node
4533 | {
4534 | "mcpServers": {
4535 | "weather": {
4536 | "command": "node",
4537 | "args": [
4538 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js"
4539 | ]
4540 | }
4541 | }
4542 | }
4543 | ```
4544 | </CodeGroup>
4545 | </Tab>
4546 |
4547 | <Tab title="Windows">
4548 | <CodeGroup>
4549 | ```json Node
4550 | {
4551 | "mcpServers": {
4552 | "weather": {
4553 | "command": "node",
4554 | "args": [
4555 | "C:\\PATH\\TO\\PARENT\\FOLDER\\weather\\build\\index.js"
4556 | ]
4557 | }
4558 | }
4559 | }
4560 | ```
4561 | </CodeGroup>
4562 | </Tab>
4563 | </Tabs>
4564 |
4565 | This tells Claude for Desktop:
4566 |
4567 | 1. There's an MCP server named "weather"
4568 | 2. Launch it by running `node /ABSOLUTE/PATH/TO/PARENT/FOLDER/weather/build/index.js`
4569 |
4570 | Save the file, and restart **Claude for Desktop**.
4571 | </Tab>
4572 |
4573 | <Tab title="Java">
4574 | <Note>
4575 | This is a quickstart demo based on Spring AI MCP auto-configuraiton and boot starters.
4576 | To learn how to create sync and async MCP Servers, manually, consult the [Java SDK Server](/sdk/java/mcp-server) documentation.
4577 | </Note>
4578 |
4579 | Let's get started with building our weather server!
4580 | [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)
4581 |
4582 | 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.
4583 | For manual MCP Server implementation, refer to the [MCP Server Java SDK documentation](/sdk/java/mcp-server).
4584 |
4585 | ### System requirements
4586 |
4587 | * Java 17 or higher installed.
4588 | * [Spring Boot 3.3.x](https://docs.spring.io/spring-boot/installing.html) or higher
4589 |
4590 | ### Set up your environment
4591 |
4592 | Use the [Spring Initizer](https://start.spring.io/) to bootstrat the project.
4593 |
4594 | You will need to add the following dependencies:
4595 |
4596 | <Tabs>
4597 | <Tab title="Maven">
4598 | ```xml
4599 | <dependencies>
4600 | <dependency>
4601 | <groupId>org.springframework.ai</groupId>
4602 | <artifactId>spring-ai-mcp-server-spring-boot-starter</artifactId>
4603 | </dependency>
4604 |
4605 | <dependency>
4606 | <groupId>org.springframework</groupId>
4607 | <artifactId>spring-web</artifactId>
4608 | </dependency>
4609 | </dependencies>
4610 | ```
4611 | </Tab>
4612 |
4613 | <Tab title="Gradle">
4614 | ```groovy
4615 | dependencies {
4616 | implementation platform("org.springframework.ai:spring-ai-mcp-server-spring-boot-starter")
4617 | implementation platform("org.springframework:spring-web")
4618 | }
4619 | ```
4620 | </Tab>
4621 | </Tabs>
4622 |
4623 | Then configure your application by setting the applicaiton properties:
4624 |
4625 | <CodeGroup>
4626 | ```bash application.properties
4627 | spring.main.bannerMode=off
4628 | logging.pattern.console=
4629 | ```
4630 |
4631 | ```yaml application.yml
4632 | logging:
4633 | pattern:
4634 | console:
4635 | spring:
4636 | main:
4637 | banner-mode: off
4638 | ```
4639 | </CodeGroup>
4640 |
4641 | 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.
4642 |
4643 | Now let's dive into building your server.
4644 |
4645 | ## Building your server
4646 |
4647 | ### Weather Service
4648 |
4649 | 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:
4650 |
4651 | ```java
4652 | @Service
4653 | public class WeatherService {
4654 |
4655 | private final RestClient restClient;
4656 |
4657 | public WeatherService() {
4658 | this.restClient = RestClient.builder()
4659 | .baseUrl("https://api.weather.gov")
4660 | .defaultHeader("Accept", "application/geo+json")
4661 | .defaultHeader("User-Agent", "WeatherApiClient/1.0 ([email protected])")
4662 | .build();
4663 | }
4664 |
4665 | @Tool(description = "Get weather forecast for a specific latitude/longitude")
4666 | public String getWeatherForecastByLocation(
4667 | double latitude, // Latitude coordinate
4668 | double longitude // Longitude coordinate
4669 | ) {
4670 | // Returns detailed forecast including:
4671 | // - Temperature and unit
4672 | // - Wind speed and direction
4673 | // - Detailed forecast description
4674 | }
4675 |
4676 | @Tool(description = "Get weather alerts for a US state")
4677 | public String getAlerts(
4678 | @ToolParam(description = "Two-letter US state code (e.g. CA, NY") String state)
4679 | ) {
4680 | // Returns active alerts including:
4681 | // - Event type
4682 | // - Affected area
4683 | // - Severity
4684 | // - Description
4685 | // - Safety instructions
4686 | }
4687 |
4688 | // ......
4689 | }
4690 | ```
4691 |
4692 | The `@Service` annotation with auto-register the service in your applicaiton context.
4693 | The Spring AI `@Tool` annotation, making it easy to create and maintain MCP tools.
4694 |
4695 | The auto-configuration will automatically register these tools with the MCP server.
4696 |
4697 | ### Create your Boot Applicaiton
4698 |
4699 | ```java
4700 | @SpringBootApplication
4701 | public class McpServerApplication {
4702 |
4703 | public static void main(String[] args) {
4704 | SpringApplication.run(McpServerApplication.class, args);
4705 | }
4706 |
4707 | @Bean
4708 | public ToolCallbackProvider weatherTools(WeatherService weatherService) {
4709 | return MethodToolCallbackProvider.builder().toolObjects(weatherService).build();
4710 | }
4711 | }
4712 | ```
4713 |
4714 | Uses the the `MethodToolCallbackProvider` utils to convert the `@Tools` into actionalble callbackes used by the MCP server.
4715 |
4716 | ### Running the server
4717 |
4718 | Finally, let's build the server:
4719 |
4720 | ```bash
4721 | ./mvnw clean install
4722 | ```
4723 |
4724 | This will generate a `mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar` file within the `target` folder.
4725 |
4726 | Let's now test your server from an existing MCP host, Claude for Desktop.
4727 |
4728 | ## Testing your server with Claude for Desktop
4729 |
4730 | <Note>
4731 | Claude for Desktop is not yet available on Linux.
4732 | </Note>
4733 |
4734 | First, make sure you have Claude for Desktop installed.
4735 | [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.**
4736 |
4737 | We'll need to configure Claude for Desktop for whichever MCP servers you want to use.
4738 | To do this, open your Claude for Desktop App configuration at `~/Library/Application Support/Claude/claude_desktop_config.json` in a text editor.
4739 | Make sure to create the file if it doesn't exist.
4740 |
4741 | For example, if you have [VS Code](https://code.visualstudio.com/) installed:
4742 |
4743 | <Tabs>
4744 | <Tab title="MacOS/Linux">
4745 | ```bash
4746 | code ~/Library/Application\ Support/Claude/claude_desktop_config.json
4747 | ```
4748 | </Tab>
4749 |
4750 | <Tab title="Windows">
4751 | ```powershell
4752 | code $env:AppData\Claude\claude_desktop_config.json
4753 | ```
4754 | </Tab>
4755 | </Tabs>
4756 |
4757 | You'll then add your servers in the `mcpServers` key.
4758 | The MCP UI elements will only show up in Claude for Desktop if at least one server is properly configured.
4759 |
4760 | In this case, we'll add our single weather server like so:
4761 |
4762 | <Tabs>
4763 | <Tab title="MacOS/Linux">
4764 | ```json java
4765 | {
4766 | "mcpServers": {
4767 | "spring-ai-mcp-weather": {
4768 | "command": "java",
4769 | "args": [
4770 | "-Dspring.ai.mcp.server.stdio=true",
4771 | "-jar",
4772 | "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4773 | ]
4774 | }
4775 | }
4776 | }
4777 | ```
4778 | </Tab>
4779 |
4780 | <Tab title="Windows">
4781 | ```json java
4782 | {
4783 | "mcpServers": {
4784 | "spring-ai-mcp-weather": {
4785 | "command": "java",
4786 | "args": [
4787 | "-Dspring.ai.mcp.server.transport=STDIO",
4788 | "-jar",
4789 | "C:\\ABSOLUTE\\PATH\\TO\\PARENT\\FOLDER\\weather\\mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar"
4790 | ]
4791 | }
4792 | }
4793 | }
4794 | ```
4795 | </Tab>
4796 | </Tabs>
4797 |
4798 | <Note>
4799 | Make sure you pass in the absolute path to your server.
4800 | </Note>
4801 |
4802 | This tells Claude for Desktop:
4803 |
4804 | 1. There's an MCP server named "my-weather-server"
4805 | 2. To launch it by running `java -jar /ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar`
4806 |
4807 | Save the file, and restart **Claude for Desktop**.
4808 |
4809 | ## Testing your server with Java client
4810 |
4811 | ### Create a MCP Client manually
4812 |
4813 | Use the `McpClient` to connect to the server:
4814 |
4815 | ```java
4816 | var stdioParams = ServerParameters.builder("java")
4817 | .args("-jar", "/ABSOLUTE/PATH/TO/PARENT/FOLDER/mcp-weather-stdio-server-0.0.1-SNAPSHOT.jar")
4818 | .build();
4819 |
4820 | var stdioTransport = new StdioClientTransport(stdioParams);
4821 |
4822 | var mcpClient = McpClient.sync(stdioTransport).build();
4823 |
4824 | mcpClient.initialize();
4825 |
4826 | ListToolsResult toolsList = mcpClient.listTools();
4827 |
4828 | CallToolResult weather = mcpClient.callTool(
4829 | new CallToolRequest("getWeatherForecastByLocation",
4830 | Map.of("latitude", "47.6062", "longitude", "-122.3321")));
4831 |
4832 | CallToolResult alert = mcpClient.callTool(
4833 | new CallToolRequest("getAlerts", Map.of("state", "NY")));
4834 |
4835 | mcpClient.closeGracefully();
4836 | ```
4837 |
4838 | ### Use MCP Client Boot Starter
4839 |
4840 | Create a new boot starter applicaiton using the `spring-ai-mcp-client-spring-boot-starter` dependency:
4841 |
4842 | ```xml
4843 | <dependency>
4844 | <groupId>org.springframework.ai</groupId>
4845 | <artifactId>spring-ai-mcp-client-spring-boot-starter</artifactId>
4846 | </dependency>
4847 | ```
4848 |
4849 | and set the `spring.ai.mcp.client.stdio.servers-configuration` property to point to your `claude_desktop_config.json`.
4850 | You can re-use the existing Anthropic Destop configuration:
4851 |
4852 | ```properties
4853 | spring.ai.mcp.client.stdio.servers-configuration=file:PATH/TO/claude_desktop_config.json
4854 | ```
4855 |
4856 | When you stasrt your client applicaiton, the auto-configuration will create, automatically MCP clients from the claude\_desktop\_config.json.
4857 |
4858 | 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.
4859 |
4860 | ## More Java MCP Server examples
4861 |
4862 | 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.
4863 | It showcases how to define and register MCP Tools, Resources, and Prompts, using the Spring Boot's auto-configuration capabilities.
4864 | </Tab>
4865 | </Tabs>
4866 |
4867 | ### Test with commands
4868 |
4869 | 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:
4870 |
4871 | <Frame>
4872 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/visual-indicator-mcp-tools.png" />
4873 | </Frame>
4874 |
4875 | After clicking on the hammer icon, you should see two tools listed:
4876 |
4877 | <Frame>
4878 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/available-mcp-tools.png" />
4879 | </Frame>
4880 |
4881 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
4882 |
4883 | If the hammer icon has shown up, you can now test your server by running the following commands in Claude for Desktop:
4884 |
4885 | * What's the weather in Sacramento?
4886 | * What are the active weather alerts in Texas?
4887 |
4888 | <Frame>
4889 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/current-weather.png" />
4890 | </Frame>
4891 |
4892 | <Frame>
4893 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/weather-alerts.png" />
4894 | </Frame>
4895 |
4896 | <Note>
4897 | Since this is the US National Weather service, the queries will only work for US locations.
4898 | </Note>
4899 |
4900 | ## What's happening under the hood
4901 |
4902 | When you ask a question:
4903 |
4904 | 1. The client sends your question to Claude
4905 | 2. Claude analyzes the available tools and decides which one(s) to use
4906 | 3. The client executes the chosen tool(s) through the MCP server
4907 | 4. The results are sent back to Claude
4908 | 5. Claude formulates a natural language response
4909 | 6. The response is displayed to you!
4910 |
4911 | ## Troubleshooting
4912 |
4913 | <AccordionGroup>
4914 | <Accordion title="Claude for Desktop Integration Issues">
4915 | **Getting logs from Claude for Desktop**
4916 |
4917 | Claude.app logging related to MCP is written to log files in `~/Library/Logs/Claude`:
4918 |
4919 | * `mcp.log` will contain general logging about MCP connections and connection failures.
4920 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
4921 |
4922 | You can run the following command to list recent logs and follow along with any new ones:
4923 |
4924 | ```bash
4925 | # Check Claude's logs for errors
4926 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
4927 | ```
4928 |
4929 | **Server not showing up in Claude**
4930 |
4931 | 1. Check your `claude_desktop_config.json` file syntax
4932 | 2. Make sure the path to your project is absolute and not relative
4933 | 3. Restart Claude for Desktop completely
4934 |
4935 | **Tool calls failing silently**
4936 |
4937 | If Claude attempts to use the tools but they fail:
4938 |
4939 | 1. Check Claude's logs for errors
4940 | 2. Verify your server builds and runs without errors
4941 | 3. Try restarting Claude for Desktop
4942 |
4943 | **None of this is working. What do I do?**
4944 |
4945 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
4946 | </Accordion>
4947 |
4948 | <Accordion title="Weather API Issues">
4949 | **Error: Failed to retrieve grid point data**
4950 |
4951 | This usually means either:
4952 |
4953 | 1. The coordinates are outside the US
4954 | 2. The NWS API is having issues
4955 | 3. You're being rate limited
4956 |
4957 | Fix:
4958 |
4959 | * Verify you're using US coordinates
4960 | * Add a small delay between requests
4961 | * Check the NWS API status page
4962 |
4963 | **Error: No active alerts for \[STATE]**
4964 |
4965 | 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.
4966 | </Accordion>
4967 | </AccordionGroup>
4968 |
4969 | <Note>
4970 | For more advanced troubleshooting, check out our guide on [Debugging MCP](/docs/tools/debugging)
4971 | </Note>
4972 |
4973 | ## Next steps
4974 |
4975 | <CardGroup cols={2}>
4976 | <Card title="Building a client" icon="outlet" href="/quickstart/client">
4977 | Learn how to build your own MCP client that can connect to your server
4978 | </Card>
4979 |
4980 | <Card title="Example servers" icon="grid" href="/examples">
4981 | Check out our gallery of official MCP servers and implementations
4982 | </Card>
4983 |
4984 | <Card title="Debugging Guide" icon="bug" href="/docs/tools/debugging">
4985 | Learn how to effectively debug MCP servers and integrations
4986 | </Card>
4987 |
4988 | <Card title="Building MCP with LLMs" icon="comments" href="/building-mcp-with-llms">
4989 | Learn how to use LLMs like Claude to speed up your MCP development
4990 | </Card>
4991 | </CardGroup>
4992 |
4993 |
4994 | # For Claude Desktop Users
4995 | Source: https://modelcontextprotocol.io/quickstart/user
4996 |
4997 | Get started using pre-built servers in Claude for Desktop.
4998 |
4999 | 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.
5000 |
5001 | <Frame>
5002 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-filesystem.png" />
5003 | </Frame>
5004 |
5005 | Don't worry — it will ask you for your permission before executing these actions!
5006 |
5007 | ## 1. Download Claude for Desktop
5008 |
5009 | Start by downloading [Claude for Desktop](https://claude.ai/download), choosing either macOS or Windows. (Linux is not yet supported for Claude for Desktop.)
5010 |
5011 | Follow the installation instructions.
5012 |
5013 | 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..."
5014 |
5015 | <Accordion title="Why Claude for Desktop and not Claude.ai?">
5016 | Because servers are locally run, MCP currently only supports desktop hosts. Remote hosts are in active development.
5017 | </Accordion>
5018 |
5019 | ## 2. Add the Filesystem MCP Server
5020 |
5021 | 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.
5022 |
5023 | 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.
5024 |
5025 | This is what it should look like on a Mac:
5026 |
5027 | <Frame style={{ textAlign: 'center' }}>
5028 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-menu.png" width="400" />
5029 | </Frame>
5030 |
5031 | Click on "Developer" in the lefthand bar of the Settings pane, and then click on "Edit Config":
5032 |
5033 | <Frame>
5034 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-developer.png" />
5035 | </Frame>
5036 |
5037 | This will create a configuration file at:
5038 |
5039 | * macOS: `~/Library/Application Support/Claude/claude_desktop_config.json`
5040 | * Windows: `%APPDATA%\Claude\claude_desktop_config.json`
5041 |
5042 | if you don't already have one, and will display the file in your file system.
5043 |
5044 | Open up the configuration file in any text editor. Replace the file contents with this:
5045 |
5046 | <Tabs>
5047 | <Tab title="MacOS/Linux">
5048 | ```json
5049 | {
5050 | "mcpServers": {
5051 | "filesystem": {
5052 | "command": "npx",
5053 | "args": [
5054 | "-y",
5055 | "@modelcontextprotocol/server-filesystem",
5056 | "/Users/username/Desktop",
5057 | "/Users/username/Downloads"
5058 | ]
5059 | }
5060 | }
5061 | }
5062 | ```
5063 | </Tab>
5064 |
5065 | <Tab title="Windows">
5066 | ```json
5067 | {
5068 | "mcpServers": {
5069 | "filesystem": {
5070 | "command": "npx",
5071 | "args": [
5072 | "-y",
5073 | "@modelcontextprotocol/server-filesystem",
5074 | "C:\\Users\\username\\Desktop",
5075 | "C:\\Users\\username\\Downloads"
5076 | ]
5077 | }
5078 | }
5079 | }
5080 | ```
5081 | </Tab>
5082 | </Tabs>
5083 |
5084 | 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.
5085 |
5086 | 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.
5087 |
5088 | * On macOS, open the Terminal from your Applications folder
5089 | * On Windows, press Windows + R, type "cmd", and press Enter
5090 |
5091 | Once in the command line, verify you have Node installed by entering in the following command:
5092 |
5093 | ```bash
5094 | node --version
5095 | ```
5096 |
5097 | If you get an error saying "command not found" or "node is not recognized", download Node from [nodejs.org](https://nodejs.org/).
5098 |
5099 | <Tip>
5100 | **How does the configuration file work?**
5101 |
5102 | 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.
5103 | </Tip>
5104 |
5105 | <Warning>
5106 | **Command Privileges**
5107 |
5108 | 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.
5109 | </Warning>
5110 |
5111 | ## 3. Restart Claude
5112 |
5113 | After updating your configuration file, you need to restart Claude for Desktop.
5114 |
5115 | 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:
5116 |
5117 | <Frame>
5118 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-hammer.png" />
5119 | </Frame>
5120 |
5121 | After clicking on the hammer icon, you should see the tools that come with the Filesystem MCP Server:
5122 |
5123 | <Frame style={{ textAlign: 'center' }}>
5124 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-tools.png" width="400" />
5125 | </Frame>
5126 |
5127 | If your server isn't being picked up by Claude for Desktop, proceed to the [Troubleshooting](#troubleshooting) section for debugging tips.
5128 |
5129 | ## 4. Try it out!
5130 |
5131 | You can now talk to Claude and ask it about your filesystem. It should know when to call the relevant tools.
5132 |
5133 | Things you might try asking Claude:
5134 |
5135 | * Can you write a poem and save it to my desktop?
5136 | * What are some work-related files in my downloads folder?
5137 | * Can you take all the images on my desktop and move them to a new folder called "Images"?
5138 |
5139 | As needed, Claude will call the relevant tools and seek your approval before taking an action:
5140 |
5141 | <Frame style={{ textAlign: 'center' }}>
5142 | <img src="https://mintlify.s3.us-west-1.amazonaws.com/mcp/images/quickstart-approve.png" width="500" />
5143 | </Frame>
5144 |
5145 | ## Troubleshooting
5146 |
5147 | <AccordionGroup>
5148 | <Accordion title="Server not showing up in Claude / hammer icon missing">
5149 | 1. Restart Claude for Desktop completely
5150 | 2. Check your `claude_desktop_config.json` file syntax
5151 | 3. Make sure the file paths included in `claude_desktop_config.json` are valid and that they are absolute and not relative
5152 | 4. Look at [logs](#getting-logs-from-claude-for-desktop) to see why the server is not connecting
5153 | 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:
5154 |
5155 | <Tabs>
5156 | <Tab title="MacOS/Linux">
5157 | ```bash
5158 | npx -y @modelcontextprotocol/server-filesystem /Users/username/Desktop /Users/username/Downloads
5159 | ```
5160 | </Tab>
5161 |
5162 | <Tab title="Windows">
5163 | ```bash
5164 | npx -y @modelcontextprotocol/server-filesystem C:\Users\username\Desktop C:\Users\username\Downloads
5165 | ```
5166 | </Tab>
5167 | </Tabs>
5168 | </Accordion>
5169 |
5170 | <Accordion title="Getting logs from Claude for Desktop">
5171 | Claude.app logging related to MCP is written to log files in:
5172 |
5173 | * macOS: `~/Library/Logs/Claude`
5174 |
5175 | * Windows: `%APPDATA%\Claude\logs`
5176 |
5177 | * `mcp.log` will contain general logging about MCP connections and connection failures.
5178 |
5179 | * Files named `mcp-server-SERVERNAME.log` will contain error (stderr) logging from the named server.
5180 |
5181 | 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):
5182 |
5183 | <Tabs>
5184 | <Tab title="MacOS/Linux">
5185 | ```bash
5186 | # Check Claude's logs for errors
5187 | tail -n 20 -f ~/Library/Logs/Claude/mcp*.log
5188 | ```
5189 | </Tab>
5190 |
5191 | <Tab title="Windows">
5192 | ```bash
5193 | type "%APPDATA%\Claude\logs\mcp*.log"
5194 | ```
5195 | </Tab>
5196 | </Tabs>
5197 | </Accordion>
5198 |
5199 | <Accordion title="Tool calls failing silently">
5200 | If Claude attempts to use the tools but they fail:
5201 |
5202 | 1. Check Claude's logs for errors
5203 | 2. Verify your server builds and runs without errors
5204 | 3. Try restarting Claude for Desktop
5205 | </Accordion>
5206 |
5207 | <Accordion title="None of this is working. What do I do?">
5208 | Please refer to our [debugging guide](/docs/tools/debugging) for better debugging tools and more detailed guidance.
5209 | </Accordion>
5210 |
5211 | <Accordion title="ENOENT error and `${APPDATA}` in paths on Windows">
5212 | If your configured server fails to load, and you see within its logs an error referring to `${APPDATA}` within a path, you may need to add the expanded value of `%APPDATA%` to your `env` key in `claude_desktop_config.json`:
5213 |
5214 | ```json
5215 | {
5216 | "brave-search": {
5217 | "command": "npx",
5218 | "args": ["-y", "@modelcontextprotocol/server-brave-search"],
5219 | "env": {
5220 | "APPDATA": "C:\\Users\\user\\AppData\\Roaming\\",
5221 | "BRAVE_API_KEY": "..."
5222 | }
5223 | }
5224 | }
5225 | ```
5226 |
5227 | With this change in place, launch Claude Desktop once again.
5228 |
5229 | <Warning>
5230 | **NPM should be installed globally**
5231 |
5232 | 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:
5233 |
5234 | ```bash
5235 | npm install -g npm
5236 | ```
5237 | </Warning>
5238 | </Accordion>
5239 | </AccordionGroup>
5240 |
5241 | ## Next steps
5242 |
5243 | <CardGroup cols={2}>
5244 | <Card title="Explore other servers" icon="grid" href="/examples">
5245 | Check out our gallery of official MCP servers and implementations
5246 | </Card>
5247 |
5248 | <Card title="Build your own server" icon="code" href="/quickstart/server">
5249 | Now build your own custom server to use in Claude for Desktop and other clients
5250 | </Card>
5251 | </CardGroup>
5252 |
5253 |
5254 | # MCP Client
5255 | Source: https://modelcontextprotocol.io/sdk/java/mcp-client
5256 |
5257 | Learn how to use the Model Context Protocol (MCP) client to interact with MCP servers
5258 |
5259 | # Model Context Protocol Client
5260 |
5261 | 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:
5262 |
5263 | * Protocol version negotiation to ensure compatibility with servers
5264 | * Capability negotiation to determine available features
5265 | * Message transport and JSON-RPC communication
5266 | * Tool discovery and execution
5267 | * Resource access and management
5268 | * Prompt system interactions
5269 | * Optional features like roots management and sampling support
5270 |
5271 | The client provides both synchronous and asynchronous APIs for flexibility in different application contexts.
5272 |
5273 | <Tabs>
5274 | <Tab title="Sync API">
5275 | ```java
5276 | // Create a sync client with custom configuration
5277 | McpSyncClient client = McpClient.sync(transport)
5278 | .requestTimeout(Duration.ofSeconds(10))
5279 | .capabilities(ClientCapabilities.builder()
5280 | .roots(true) // Enable roots capability
5281 | .sampling() // Enable sampling capability
5282 | .build())
5283 | .sampling(request -> new CreateMessageResult(response))
5284 | .build();
5285 |
5286 | // Initialize connection
5287 | client.initialize();
5288 |
5289 | // List available tools
5290 | ListToolsResult tools = client.listTools();
5291 |
5292 | // Call a tool
5293 | CallToolResult result = client.callTool(
5294 | new CallToolRequest("calculator",
5295 | Map.of("operation", "add", "a", 2, "b", 3))
5296 | );
5297 |
5298 | // List and read resources
5299 | ListResourcesResult resources = client.listResources();
5300 | ReadResourceResult resource = client.readResource(
5301 | new ReadResourceRequest("resource://uri")
5302 | );
5303 |
5304 | // List and use prompts
5305 | ListPromptsResult prompts = client.listPrompts();
5306 | GetPromptResult prompt = client.getPrompt(
5307 | new GetPromptRequest("greeting", Map.of("name", "Spring"))
5308 | );
5309 |
5310 | // Add/remove roots
5311 | client.addRoot(new Root("file:///path", "description"));
5312 | client.removeRoot("file:///path");
5313 |
5314 | // Close client
5315 | client.closeGracefully();
5316 | ```
5317 | </Tab>
5318 |
5319 | <Tab title="Async API">
5320 | ```java
5321 | // Create an async client with custom configuration
5322 | McpAsyncClient client = McpClient.async(transport)
5323 | .requestTimeout(Duration.ofSeconds(10))
5324 | .capabilities(ClientCapabilities.builder()
5325 | .roots(true) // Enable roots capability
5326 | .sampling() // Enable sampling capability
5327 | .build())
5328 | .sampling(request -> Mono.just(new CreateMessageResult(response)))
5329 | .toolsChangeConsumer(tools -> Mono.fromRunnable(() -> {
5330 | logger.info("Tools updated: {}", tools);
5331 | }))
5332 | .resourcesChangeConsumer(resources -> Mono.fromRunnable(() -> {
5333 | logger.info("Resources updated: {}", resources);
5334 | }))
5335 | .promptsChangeConsumer(prompts -> Mono.fromRunnable(() -> {
5336 | logger.info("Prompts updated: {}", prompts);
5337 | }))
5338 | .build();
5339 |
5340 | // Initialize connection and use features
5341 | client.initialize()
5342 | .flatMap(initResult -> client.listTools())
5343 | .flatMap(tools -> {
5344 | return client.callTool(new CallToolRequest(
5345 | "calculator",
5346 | Map.of("operation", "add", "a", 2, "b", 3)
5347 | ));
5348 | })
5349 | .flatMap(result -> {
5350 | return client.listResources()
5351 | .flatMap(resources ->
5352 | client.readResource(new ReadResourceRequest("resource://uri"))
5353 | );
5354 | })
5355 | .flatMap(resource -> {
5356 | return client.listPrompts()
5357 | .flatMap(prompts ->
5358 | client.getPrompt(new GetPromptRequest(
5359 | "greeting",
5360 | Map.of("name", "Spring")
5361 | ))
5362 | );
5363 | })
5364 | .flatMap(prompt -> {
5365 | return client.addRoot(new Root("file:///path", "description"))
5366 | .then(client.removeRoot("file:///path"));
5367 | })
5368 | .doFinally(signalType -> {
5369 | client.closeGracefully().subscribe();
5370 | })
5371 | .subscribe();
5372 | ```
5373 | </Tab>
5374 | </Tabs>
5375 |
5376 | ## Client Transport
5377 |
5378 | 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.
5379 |
5380 | <Tabs>
5381 | <Tab title="STDIO">
5382 | Creates transport for in-process based communication
5383 |
5384 | ```java
5385 | ServerParameters params = ServerParameters.builder("npx")
5386 | .args("-y", "@modelcontextprotocol/server-everything", "dir")
5387 | .build();
5388 | McpTransport transport = new StdioClientTransport(params);
5389 | ```
5390 | </Tab>
5391 |
5392 | <Tab title="SSE (HttpClient)">
5393 | Creates a framework agnostic (pure Java API) SSE client transport. Included in the core mcp module.
5394 |
5395 | ```java
5396 | McpTransport transport = new HttpClientSseClientTransport("http://your-mcp-server");
5397 | ```
5398 | </Tab>
5399 |
5400 | <Tab title="SSE (WebFlux)">
5401 | Creates WebFlux-based SSE client transport. Requires the mcp-webflux-sse-transport dependency.
5402 |
5403 | ```java
5404 | WebClient.Builder webClientBuilder = WebClient.builder()
5405 | .baseUrl("http://your-mcp-server");
5406 | McpTransport transport = new WebFluxSseClientTransport(webClientBuilder);
5407 | ```
5408 | </Tab>
5409 | </Tabs>
5410 |
5411 | ## Client Capabilities
5412 |
5413 | The client can be configured with various capabilities:
5414 |
5415 | ```java
5416 | var capabilities = ClientCapabilities.builder()
5417 | .roots(true) // Enable filesystem roots support with list changes notifications
5418 | .sampling() // Enable LLM sampling support
5419 | .build();
5420 | ```
5421 |
5422 | ### Roots Support
5423 |
5424 | Roots define the boundaries of where servers can operate within the filesystem:
5425 |
5426 | ```java
5427 | // Add a root dynamically
5428 | client.addRoot(new Root("file:///path", "description"));
5429 |
5430 | // Remove a root
5431 | client.removeRoot("file:///path");
5432 |
5433 | // Notify server of roots changes
5434 | client.rootsListChangedNotification();
5435 | ```
5436 |
5437 | The roots capability allows servers to:
5438 |
5439 | * Request the list of accessible filesystem roots
5440 | * Receive notifications when the roots list changes
5441 | * Understand which directories and files they have access to
5442 |
5443 | ### Sampling Support
5444 |
5445 | Sampling enables servers to request LLM interactions ("completions" or "generations") through the client:
5446 |
5447 | ```java
5448 | // Configure sampling handler
5449 | Function<CreateMessageRequest, CreateMessageResult> samplingHandler = request -> {
5450 | // Sampling implementation that interfaces with LLM
5451 | return new CreateMessageResult(response);
5452 | };
5453 |
5454 | // Create client with sampling support
5455 | var client = McpClient.sync(transport)
5456 | .capabilities(ClientCapabilities.builder()
5457 | .sampling()
5458 | .build())
5459 | .sampling(samplingHandler)
5460 | .build();
5461 | ```
5462 |
5463 | This capability allows:
5464 |
5465 | * Servers to leverage AI capabilities without requiring API keys
5466 | * Clients to maintain control over model access and permissions
5467 | * Support for both text and image-based interactions
5468 | * Optional inclusion of MCP server context in prompts
5469 |
5470 | ## Using MCP Clients
5471 |
5472 | ### Tool Execution
5473 |
5474 | 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.
5475 |
5476 | <Tabs>
5477 | <Tab title="Sync API">
5478 | ```java
5479 | // List available tools and their names
5480 | var tools = client.listTools();
5481 | tools.forEach(tool -> System.out.println(tool.getName()));
5482 |
5483 | // Execute a tool with parameters
5484 | var result = client.callTool("calculator", Map.of(
5485 | "operation", "add",
5486 | "a", 1,
5487 | "b", 2
5488 | ));
5489 | ```
5490 | </Tab>
5491 |
5492 | <Tab title="Async API">
5493 | ```java
5494 | // List available tools asynchronously
5495 | client.listTools()
5496 | .doOnNext(tools -> tools.forEach(tool ->
5497 | System.out.println(tool.getName())))
5498 | .subscribe();
5499 |
5500 | // Execute a tool asynchronously
5501 | client.callTool("calculator", Map.of(
5502 | "operation", "add",
5503 | "a", 1,
5504 | "b", 2
5505 | ))
5506 | .subscribe();
5507 | ```
5508 | </Tab>
5509 | </Tabs>
5510 |
5511 | ### Resource Access
5512 |
5513 | 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.
5514 |
5515 | <Tabs>
5516 | <Tab title="Sync API">
5517 | ```java
5518 | // List available resources and their names
5519 | var resources = client.listResources();
5520 | resources.forEach(resource -> System.out.println(resource.getName()));
5521 |
5522 | // Retrieve resource content using a URI template
5523 | var content = client.getResource("file", Map.of(
5524 | "path", "/path/to/file.txt"
5525 | ));
5526 | ```
5527 | </Tab>
5528 |
5529 | <Tab title="Async API">
5530 | ```java
5531 | // List available resources asynchronously
5532 | client.listResources()
5533 | .doOnNext(resources -> resources.forEach(resource ->
5534 | System.out.println(resource.getName())))
5535 | .subscribe();
5536 |
5537 | // Retrieve resource content asynchronously
5538 | client.getResource("file", Map.of(
5539 | "path", "/path/to/file.txt"
5540 | ))
5541 | .subscribe();
5542 | ```
5543 | </Tab>
5544 | </Tabs>
5545 |
5546 | ### Prompt System
5547 |
5548 | 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.
5549 |
5550 | <Tabs>
5551 | <Tab title="Sync API">
5552 | ```java
5553 | // List available prompt templates
5554 | var prompts = client.listPrompts();
5555 | prompts.forEach(prompt -> System.out.println(prompt.getName()));
5556 |
5557 | // Execute a prompt template with parameters
5558 | var response = client.executePrompt("echo", Map.of(
5559 | "text", "Hello, World!"
5560 | ));
5561 | ```
5562 | </Tab>
5563 |
5564 | <Tab title="Async API">
5565 | ```java
5566 | // List available prompt templates asynchronously
5567 | client.listPrompts()
5568 | .doOnNext(prompts -> prompts.forEach(prompt ->
5569 | System.out.println(prompt.getName())))
5570 | .subscribe();
5571 |
5572 | // Execute a prompt template asynchronously
5573 | client.executePrompt("echo", Map.of(
5574 | "text", "Hello, World!"
5575 | ))
5576 | .subscribe();
5577 | ```
5578 | </Tab>
5579 | </Tabs>
5580 |
5581 |
5582 | # Overview
5583 | Source: https://modelcontextprotocol.io/sdk/java/mcp-overview
5584 |
5585 | Introduction to the Model Context Protocol (MCP) Java SDK
5586 |
5587 | Java SDK for the [Model Context Protocol](https://modelcontextprotocol.org/docs/concepts/architecture)
5588 | enables standardized integration between AI models and tools.
5589 |
5590 | ## Features
5591 |
5592 | * MCP Client and MCP Server implementations supporting:
5593 | * Protocol [version compatibility negotiation](https://spec.modelcontextprotocol.io/specification/2024-11-05/basic/lifecycle/#initialization)
5594 | * [Tool](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/tools/) discovery, execution, list change notifications
5595 | * [Resource](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/resources/) management with URI templates
5596 | * [Roots](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/roots/) list management and notifications
5597 | * [Prompt](https://spec.modelcontextprotocol.io/specification/2024-11-05/server/prompts/) handling and management
5598 | * [Sampling](https://spec.modelcontextprotocol.io/specification/2024-11-05/client/sampling/) support for AI model interactions
5599 | * Multiple transport implementations:
5600 | * Default transports:
5601 | * Stdio-based transport for process-based communication
5602 | * Java HttpClient-based SSE client transport for HTTP SSE Client-side streaming
5603 | * Servlet-based SSE server transport for HTTP SSE Server streaming
5604 | * Spring-based transports:
5605 | * WebFlux SSE client and server transports for reactive HTTP streaming
5606 | * WebMVC SSE transport for servlet-based HTTP streaming
5607 | * Supports Synchronous and Asynchronous programming paradigms
5608 |
5609 | ## Architecture
5610 |
5611 | The SDK follows a layered architecture with clear separation of concerns:
5612 |
5613 | 
5614 |
5615 | * **Client/Server Layer (McpClient/McpServer)**: Both use McpSession for sync/async operations,
5616 | with McpClient handling client-side protocol operations and McpServer managing server-side protocol operations.
5617 | * **Session Layer (McpSession)**: Manages communication patterns and state using DefaultMcpSession implementation.
5618 | * **Transport Layer (McpTransport)**: Handles JSON-RPC message serialization/deserialization via:
5619 | * StdioTransport (stdin/stdout) in the core module
5620 | * HTTP SSE transports in dedicated transport modules (Java HttpClient, Spring WebFlux, Spring WebMVC)
5621 |
5622 | The MCP Client is a key component in the Model Context Protocol (MCP) architecture, responsible for establishing and managing connections with MCP servers.
5623 | It implements the client-side of the protocol.
5624 |
5625 | 
5626 |
5627 | The MCP Server is a foundational component in the Model Context Protocol (MCP) architecture that provides tools, resources, and capabilities to clients.
5628 | It implements the server-side of the protocol.
5629 |
5630 | 
5631 |
5632 | Key Interactions:
5633 |
5634 | * **Client/Server Initialization**: Transport setup, protocol compatibility check, capability negotiation, and implementation details exchange.
5635 | * **Message Flow**: JSON-RPC message handling with validation, type-safe response processing, and error handling.
5636 | * **Resource Management**: Resource discovery, URI template-based access, subscription system, and content retrieval.
5637 |
5638 | ## Dependencies
5639 |
5640 | Add the following Maven dependency to your project:
5641 |
5642 | <Tabs>
5643 | <Tab title="Maven">
5644 | The core MCP functionality:
5645 |
5646 | ```xml
5647 | <dependency>
5648 | <groupId>io.modelcontextprotocol.sdk</groupId>
5649 | <artifactId>mcp</artifactId>
5650 | </dependency>
5651 | ```
5652 |
5653 | For HTTP SSE transport implementations, add one of the following dependencies:
5654 |
5655 | ```xml
5656 | <!-- Spring WebFlux-based SSE client and server transport -->
5657 | <dependency>
5658 | <groupId>io.modelcontextprotocol.sdk</groupId>
5659 | <artifactId>mcp-spring-webflux</artifactId>
5660 | </dependency>
5661 |
5662 | <!-- Spring WebMVC-based SSE server transport -->
5663 | <dependency>
5664 | <groupId>io.modelcontextprotocol.sdk</groupId>
5665 | <artifactId>mcp-spring-webmvc</artifactId>
5666 | </dependency>
5667 | ```
5668 | </Tab>
5669 |
5670 | <Tab title="Gradle">
5671 | The core MCP functionality:
5672 |
5673 | ```groovy
5674 | dependencies {
5675 | implementation platform("io.modelcontextprotocol.sdk:mcp")
5676 | //...
5677 | }
5678 | ```
5679 |
5680 | For HTTP SSE transport implementations, add one of the following dependencies:
5681 |
5682 | ```groovy
5683 | // Spring WebFlux-based SSE client and server transport
5684 | dependencies {
5685 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webflux")
5686 | }
5687 |
5688 | // Spring WebMVC-based SSE server transport
5689 | dependencies {
5690 | implementation platform("io.modelcontextprotocol.sdk:mcp-spring-webmvc")
5691 | }
5692 | ```
5693 | </Tab>
5694 | </Tabs>
5695 |
5696 | ### Bill of Materials (BOM)
5697 |
5698 | The Bill of Materials (BOM) declares the recommended versions of all the dependencies used by a given release.
5699 | Using the BOM from your application's build script avoids the need for you to specify and maintain the dependency versions yourself.
5700 | Instead, the version of the BOM you're using determines the utilized dependency versions.
5701 | It also ensures that you're using supported and tested versions of the dependencies by default, unless you choose to override them.
5702 |
5703 | Add the BOM to your project:
5704 |
5705 | <Tabs>
5706 | <Tab title="Maven">
5707 | ```xml
5708 | <dependencyManagement>
5709 | <dependencies>
5710 | <dependency>
5711 | <groupId>io.modelcontextprotocol.sdk</groupId>
5712 | <artifactId>mcp-bom</artifactId>
5713 | <version>0.7.0</version>
5714 | <type>pom</type>
5715 | <scope>import</scope>
5716 | </dependency>
5717 | </dependencies>
5718 | </dependencyManagement>
5719 | ```
5720 | </Tab>
5721 |
5722 | <Tab title="Gradle">
5723 | ```groovy
5724 | dependencies {
5725 | implementation platform("io.modelcontextprotocol.sdk:mcp-bom:0.7.0")
5726 | //...
5727 | }
5728 | ```
5729 |
5730 | 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.
5731 | This is implemented by adding a 'platform' dependency handler method to the dependencies section of your Gradle build script.
5732 | 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.
5733 | </Tab>
5734 | </Tabs>
5735 |
5736 | Replace the version number with the version of the BOM you want to use.
5737 |
5738 | ### Available Dependencies
5739 |
5740 | The following dependencies are available and managed by the BOM:
5741 |
5742 | * Core Dependencies
5743 | * `io.modelcontextprotocol.sdk:mcp` - Core MCP library providing the base functionality and APIs for Model Context Protocol implementation.
5744 | * Transport Dependencies
5745 | * `io.modelcontextprotocol.sdk:mcp-spring-webflux` - WebFlux-based Server-Sent Events (SSE) transport implementation for reactive applications.
5746 | * `io.modelcontextprotocol.sdk:mcp-spring-webmvc` - WebMVC-based Server-Sent Events (SSE) transport implementation for servlet-based applications.
5747 | * Testing Dependencies
5748 | * `io.modelcontextprotocol.sdk:mcp-test` - Testing utilities and support for MCP-based applications.
5749 |
5750 |
5751 | # MCP Server
5752 | Source: https://modelcontextprotocol.io/sdk/java/mcp-server
5753 |
5754 | Learn how to implement and configure a Model Context Protocol (MCP) server
5755 |
5756 | ## Overview
5757 |
5758 | 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:
5759 |
5760 | * Exposing tools that clients can discover and execute
5761 | * Managing resources with URI-based access patterns
5762 | * Providing prompt templates and handling prompt requests
5763 | * Supporting capability negotiation with clients
5764 | * Implementing server-side protocol operations
5765 | * Managing concurrent client connections
5766 | * Providing structured logging and notifications
5767 |
5768 | The server supports both synchronous and asynchronous APIs, allowing for flexible integration in different application contexts.
5769 |
5770 | <Tabs>
5771 | <Tab title="Sync API">
5772 | ```java
5773 | // Create a server with custom configuration
5774 | McpSyncServer syncServer = McpServer.sync(transport)
5775 | .serverInfo("my-server", "1.0.0")
5776 | .capabilities(ServerCapabilities.builder()
5777 | .resources(true) // Enable resource support
5778 | .tools(true) // Enable tool support
5779 | .prompts(true) // Enable prompt support
5780 | .logging() // Enable logging support
5781 | .build())
5782 | .build();
5783 |
5784 | // Initialize the server
5785 | syncServer.initialize();
5786 |
5787 | // Register tools, resources, and prompts
5788 | syncServer.addTool(syncToolRegistration);
5789 | syncServer.addResource(syncResourceRegistration);
5790 | syncServer.addPrompt(syncPromptRegistration);
5791 |
5792 | // Send logging notifications
5793 | syncServer.loggingNotification(LoggingMessageNotification.builder()
5794 | .level(LoggingLevel.INFO)
5795 | .logger("custom-logger")
5796 | .data("Server initialized")
5797 | .build());
5798 |
5799 | // Close the server when done
5800 | syncServer.close();
5801 | ```
5802 | </Tab>
5803 |
5804 | <Tab title="Async API">
5805 | ```java
5806 | // Create an async server with custom configuration
5807 | McpAsyncServer asyncServer = McpServer.async(transport)
5808 | .serverInfo("my-server", "1.0.0")
5809 | .capabilities(ServerCapabilities.builder()
5810 | .resources(true) // Enable resource support
5811 | .tools(true) // Enable tool support
5812 | .prompts(true) // Enable prompt support
5813 | .logging() // Enable logging support
5814 | .build())
5815 | .build();
5816 |
5817 | // Initialize the server
5818 | asyncServer.initialize()
5819 | .doOnSuccess(v -> logger.info("Server initialized"))
5820 | .subscribe();
5821 |
5822 | // Register tools, resources, and prompts
5823 | asyncServer.addTool(asyncToolRegistration)
5824 | .doOnSuccess(v -> logger.info("Tool registered"))
5825 | .subscribe();
5826 |
5827 | asyncServer.addResource(asyncResourceRegistration)
5828 | .doOnSuccess(v -> logger.info("Resource registered"))
5829 | .subscribe();
5830 |
5831 | asyncServer.addPrompt(asyncPromptRegistration)
5832 | .doOnSuccess(v -> logger.info("Prompt registered"))
5833 | .subscribe();
5834 |
5835 | // Send logging notifications
5836 | asyncServer.loggingNotification(LoggingMessageNotification.builder()
5837 | .level(LoggingLevel.INFO)
5838 | .logger("custom-logger")
5839 | .data("Server initialized")
5840 | .build());
5841 |
5842 | // Close the server when done
5843 | asyncServer.close()
5844 | .doOnSuccess(v -> logger.info("Server closed"))
5845 | .subscribe();
5846 | ```
5847 | </Tab>
5848 | </Tabs>
5849 |
5850 | ## Server Transport
5851 |
5852 | 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:
5853 |
5854 | <Tabs>
5855 | <Tab title="STDIO">
5856 | <>
5857 | Create in-process based transport:
5858 |
5859 | ```java
5860 | StdioServerTransport transport = new StdioServerTransport(new ObjectMapper());
5861 | ```
5862 |
5863 | Provides bidirectional JSON-RPC message handling over standard input/output streams with non-blocking message processing, serialization/deserialization, and graceful shutdown support.
5864 |
5865 | Key features:
5866 |
5867 | <ul>
5868 | <li>Bidirectional communication through stdin/stdout</li>
5869 | <li>Process-based integration support</li>
5870 | <li>Simple setup and configuration</li>
5871 | <li>Lightweight implementation</li>
5872 | </ul>
5873 | </>
5874 | </Tab>
5875 |
5876 | <Tab title="SSE (WebFlux)">
5877 | <>
5878 | <p>Creates WebFlux-based SSE server transport.<br />Requires the <code>mcp-spring-webflux</code> dependency.</p>
5879 |
5880 | ```java
5881 | @Configuration
5882 | class McpConfig {
5883 | @Bean
5884 | WebFluxSseServerTransport webFluxSseServerTransport(ObjectMapper mapper) {
5885 | return new WebFluxSseServerTransport(mapper, "/mcp/message");
5886 | }
5887 |
5888 | @Bean
5889 | RouterFunction<?> mcpRouterFunction(WebFluxSseServerTransport transport) {
5890 | return transport.getRouterFunction();
5891 | }
5892 | }
5893 | ```
5894 |
5895 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5896 |
5897 | <ul>
5898 | <li>Reactive HTTP streaming with WebFlux</li>
5899 | <li>Concurrent client connections through SSE endpoints</li>
5900 | <li>Message routing and session management</li>
5901 | <li>Graceful shutdown capabilities</li>
5902 | </ul>
5903 | </>
5904 | </Tab>
5905 |
5906 | <Tab title="SSE (WebMvc)">
5907 | <>
5908 | <p>Creates WebMvc-based SSE server transport.<br />Requires the <code>mcp-spring-webmvc</code> dependency.</p>
5909 |
5910 | ```java
5911 | @Configuration
5912 | @EnableWebMvc
5913 | class McpConfig {
5914 | @Bean
5915 | WebMvcSseServerTransport webMvcSseServerTransport(ObjectMapper mapper) {
5916 | return new WebMvcSseServerTransport(mapper, "/mcp/message");
5917 | }
5918 |
5919 | @Bean
5920 | RouterFunction<ServerResponse> mcpRouterFunction(WebMvcSseServerTransport transport) {
5921 | return transport.getRouterFunction();
5922 | }
5923 | }
5924 | ```
5925 |
5926 | <p>Implements the MCP HTTP with SSE transport specification, providing:</p>
5927 |
5928 | <ul>
5929 | <li>Server-side event streaming</li>
5930 | <li>Integration with Spring WebMVC</li>
5931 | <li>Support for traditional web applications</li>
5932 | <li>Synchronous operation handling</li>
5933 | </ul>
5934 | </>
5935 | </Tab>
5936 |
5937 | <Tab title="SSE (Servlet)">
5938 | <>
5939 | <p>
5940 | Creates a Servlet-based SSE server transport. It is included in the core <code>mcp</code> module.<br />
5941 | The <code>HttpServletSseServerTransport</code> can be used with any Servlet container.<br />
5942 | To use it with a Spring Web application, you can register it as a Servlet bean:
5943 | </p>
5944 |
5945 | ```java
5946 | @Configuration
5947 | @EnableWebMvc
5948 | public class McpServerConfig implements WebMvcConfigurer {
5949 |
5950 | @Bean
5951 | public HttpServletSseServerTransport servletSseServerTransport() {
5952 | return new HttpServletSseServerTransport(new ObjectMapper(), "/mcp/message");
5953 | }
5954 |
5955 | @Bean
5956 | public ServletRegistrationBean customServletBean(HttpServletSseServerTransport servlet) {
5957 | return new ServletRegistrationBean(servlet);
5958 | }
5959 | }
5960 | ```
5961 |
5962 | <p>
5963 | Implements the MCP HTTP with SSE transport specification using the traditional Servlet API, providing:
5964 | </p>
5965 |
5966 | <ul>
5967 | <li>Asynchronous message handling using Servlet 6.0 async support</li>
5968 | <li>Session management for multiple client connections</li>
5969 |
5970 | <li>
5971 | Two types of endpoints:
5972 |
5973 | <ul>
5974 | <li>SSE endpoint (<code>/sse</code>) for server-to-client events</li>
5975 | <li>Message endpoint (configurable) for client-to-server requests</li>
5976 | </ul>
5977 | </li>
5978 |
5979 | <li>Error handling and response formatting</li>
5980 | <li>Graceful shutdown support</li>
5981 | </ul>
5982 | </>
5983 | </Tab>
5984 | </Tabs>
5985 |
5986 | ## Server Capabilities
5987 |
5988 | The server can be configured with various capabilities:
5989 |
5990 | ```java
5991 | var capabilities = ServerCapabilities.builder()
5992 | .resources(false, true) // Resource support with list changes notifications
5993 | .tools(true) // Tool support with list changes notifications
5994 | .prompts(true) // Prompt support with list changes notifications
5995 | .logging() // Enable logging support (enabled by default with loging level INFO)
5996 | .build();
5997 | ```
5998 |
5999 | ### Logging Support
6000 |
6001 | The server provides structured logging capabilities that allow sending log messages to clients with different severity levels:
6002 |
6003 | ```java
6004 | // Send a log message to clients
6005 | server.loggingNotification(LoggingMessageNotification.builder()
6006 | .level(LoggingLevel.INFO)
6007 | .logger("custom-logger")
6008 | .data("Custom log message")
6009 | .build());
6010 | ```
6011 |
6012 | Clients can control the minimum logging level they receive through the `mcpClient.setLoggingLevel(level)` request. Messages below the set level will be filtered out.
6013 | Supported logging levels (in order of increasing severity): DEBUG (0), INFO (1), NOTICE (2), WARNING (3), ERROR (4), CRITICAL (5), ALERT (6), EMERGENCY (7)
6014 |
6015 | ### Tool Registration
6016 |
6017 | <Tabs>
6018 | <Tab title="Sync">
6019 | ```java
6020 | // Sync tool registration
6021 | var syncToolRegistration = new McpServerFeatures.SyncToolRegistration(
6022 | new Tool("calculator", "Basic calculator", Map.of(
6023 | "operation", "string",
6024 | "a", "number",
6025 | "b", "number"
6026 | )),
6027 | arguments -> {
6028 | // Tool implementation
6029 | return new CallToolResult(result, false);
6030 | }
6031 | );
6032 | ```
6033 | </Tab>
6034 |
6035 | <Tab title="Async">
6036 | ```java
6037 | // Async tool registration
6038 | var asyncToolRegistration = new McpServerFeatures.AsyncToolRegistration(
6039 | new Tool("calculator", "Basic calculator", Map.of(
6040 | "operation", "string",
6041 | "a", "number",
6042 | "b", "number"
6043 | )),
6044 | arguments -> {
6045 | // Tool implementation
6046 | return Mono.just(new CallToolResult(result, false));
6047 | }
6048 | );
6049 | ```
6050 | </Tab>
6051 | </Tabs>
6052 |
6053 | ### Resource Registration
6054 |
6055 | <Tabs>
6056 | <Tab title="Sync">
6057 | ```java
6058 | // Sync resource registration
6059 | var syncResourceRegistration = new McpServerFeatures.SyncResourceRegistration(
6060 | new Resource("custom://resource", "name", "description", "mime-type", null),
6061 | request -> {
6062 | // Resource read implementation
6063 | return new ReadResourceResult(contents);
6064 | }
6065 | );
6066 | ```
6067 | </Tab>
6068 |
6069 | <Tab title="Async">
6070 | ```java
6071 | // Async resource registration
6072 | var asyncResourceRegistration = new McpServerFeatures.AsyncResourceRegistration(
6073 | new Resource("custom://resource", "name", "description", "mime-type", null),
6074 | request -> {
6075 | // Resource read implementation
6076 | return Mono.just(new ReadResourceResult(contents));
6077 | }
6078 | );
6079 | ```
6080 | </Tab>
6081 | </Tabs>
6082 |
6083 | ### Prompt Registration
6084 |
6085 | <Tabs>
6086 | <Tab title="Sync">
6087 | ```java
6088 | // Sync prompt registration
6089 | var syncPromptRegistration = new McpServerFeatures.SyncPromptRegistration(
6090 | new Prompt("greeting", "description", List.of(
6091 | new PromptArgument("name", "description", true)
6092 | )),
6093 | request -> {
6094 | // Prompt implementation
6095 | return new GetPromptResult(description, messages);
6096 | }
6097 | );
6098 | ```
6099 | </Tab>
6100 |
6101 | <Tab title="Async">
6102 | ```java
6103 | // Async prompt registration
6104 | var asyncPromptRegistration = new McpServerFeatures.AsyncPromptRegistration(
6105 | new Prompt("greeting", "description", List.of(
6106 | new PromptArgument("name", "description", true)
6107 | )),
6108 | request -> {
6109 | // Prompt implementation
6110 | return Mono.just(new GetPromptResult(description, messages));
6111 | }
6112 | );
6113 | ```
6114 | </Tab>
6115 | </Tabs>
6116 |
6117 | ## Error Handling
6118 |
6119 | 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.
6120 |
6121 |
6122 | # Building MCP with LLMs
6123 | Source: https://modelcontextprotocol.io/tutorials/building-mcp-with-llms
6124 |
6125 | Speed up your MCP development using LLMs such as Claude!
6126 |
6127 | 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.
6128 |
6129 | ## Preparing the documentation
6130 |
6131 | Before starting, gather the necessary documentation to help Claude understand MCP:
6132 |
6133 | 1. Visit [https://modelcontextprotocol.io/llms-full.txt](https://modelcontextprotocol.io/llms-full.txt) and copy the full documentation text
6134 | 2. Navigate to either the [MCP TypeScript SDK](https://github.com/modelcontextprotocol/typescript-sdk) or [Python SDK repository](https://github.com/modelcontextprotocol/python-sdk)
6135 | 3. Copy the README files and other relevant documentation
6136 | 4. Paste these documents into your conversation with Claude
6137 |
6138 | ## Describing your server
6139 |
6140 | Once you've provided the documentation, clearly describe to Claude what kind of server you want to build. Be specific about:
6141 |
6142 | * What resources your server will expose
6143 | * What tools it will provide
6144 | * Any prompts it should offer
6145 | * What external systems it needs to interact with
6146 |
6147 | For example:
6148 |
6149 | ```
6150 | Build an MCP server that:
6151 | - Connects to my company's PostgreSQL database
6152 | - Exposes table schemas as resources
6153 | - Provides tools for running read-only SQL queries
6154 | - Includes prompts for common data analysis tasks
6155 | ```
6156 |
6157 | ## Working with Claude
6158 |
6159 | When working with Claude on MCP servers:
6160 |
6161 | 1. Start with the core functionality first, then iterate to add more features
6162 | 2. Ask Claude to explain any parts of the code you don't understand
6163 | 3. Request modifications or improvements as needed
6164 | 4. Have Claude help you test the server and handle edge cases
6165 |
6166 | Claude can help implement all the key MCP features:
6167 |
6168 | * Resource management and exposure
6169 | * Tool definitions and implementations
6170 | * Prompt templates and handlers
6171 | * Error handling and logging
6172 | * Connection and transport setup
6173 |
6174 | ## Best practices
6175 |
6176 | When building MCP servers with Claude:
6177 |
6178 | * Break down complex servers into smaller pieces
6179 | * Test each component thoroughly before moving on
6180 | * Keep security in mind - validate inputs and limit access appropriately
6181 | * Document your code well for future maintenance
6182 | * Follow MCP protocol specifications carefully
6183 |
6184 | ## Next steps
6185 |
6186 | After Claude helps you build your server:
6187 |
6188 | 1. Review the generated code carefully
6189 | 2. Test the server with the MCP Inspector tool
6190 | 3. Connect it to Claude.app or other MCP clients
6191 | 4. Iterate based on real usage and feedback
6192 |
6193 | Remember that Claude can help you modify and improve your server as requirements change over time.
6194 |
6195 | Need more guidance? Just ask Claude specific questions about implementing MCP features or troubleshooting issues that arise.
6196 |
6197 |
6198 | Saved.
```