Model Context Protocol (MCP) provides a standardized interface for Large Language Models (LLMs) to interact with external data sources and tools, acting as a universal adapter for AI applications. This guide focuses on developing MCP servers and clients, particularly for tool integration, using Python 3.11 and uv for project management.

How It Works

MCP defines a protocol for LLMs to discover and invoke tools. Servers expose tools via decorators (@app.tool()), specifying function signatures, descriptions, and argument schemas. Clients can then discover these tools and execute them. MCP supports stdio and SSE transport protocols for communication, with stdio being the primary focus for local development and SSE for serverless deployments.

Quick Start & Requirements

• Server Setup:

  uv init mcp_getting_started
  cd mcp_getting_started
  uv venv
  .venv\Scripts\activate.bat
  uv add "mcp[cli]" httpx openai
  

  Create web_search.py with the provided server code.
• Server Execution:

  uv run web_search.py
  

  Requires Python 3.11+, uv, httpx, openai.
• Debugging: Use npx -y @modelcontextprotocol/inspector uv run web_search.py or mcp dev web_search.py.
• Client Example: See the provided Python client code for stdio_client and ClientSession usage.
• Official Resources:
  • uv docs: https://docs.astral.sh/uv/
  • MCP Inspector: https://github.com/modelcontextprotocol/inspector
  • Langchain MCP Adapters: https://github.com/langchain-ai/langchain-mcp-adapters
  • GitHub Repo: https://github.com/liaokongVFX/MCP-Chinese-Getting-Started-Guide

Highlighted Details

• Tool Integration: Demonstrates creating a web search tool using httpx and the Zhipu AI API.
• Client-LLM Interaction: Shows how to use DeepSeek with MCP to call tools based on LLM output.
• Sampling: Enables pre- and post-execution hooks for tools, allowing user confirmation.
• Claude Desktop Integration: Explains how to configure custom MCP servers for the Claude desktop application.
• Serverless Deployment: Details deploying MCP services using SSE transport on Alibaba Cloud Function Compute.

Maintenance & Community

• The project is actively developed, with examples covering various use cases.
• Links to official documentation and community resources are provided within the README.

Licensing & Compatibility

• The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

• The guide mentions potential encoding bugs with gradio_client when using Hugging Face Spaces, recommending httpx instead.
• Claude Desktop's current support for wildcard resource paths (greeting://{name}) is noted as limited, though client-side usage is supported.
• The web-search-pro API used in examples is noted to have transitioned from free to a paid model.