MCP (Model Context Protocol) Implementation

This project provides implementations of MCP (Model Context Protocol) servers and clients in Python.

Overview

The Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to LLMs. This project includes:

1. A basic socket-based implementation (for learning purposes)
2. An official MCP SDK implementation with HTTP/SSE transport
3. LLM integration example using Claude with HTTP/SSE transport
4. FastMCP implementation - a higher-level wrapper for easier MCP server creation

Requirements

• Python 3.8 or higher
• Dependencies listed in requirements.txt

Installation

快速安装（推荐）

使用提供的安装脚本快速设置环境：

./setup.sh

这个脚本会自动：

1. 创建虚拟环境
2. 安装所有依赖
3. 创建 .env 文件（如果不存在）
4. 使所有 Python 文件可执行

手动安装

1. Clone this repository
2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

   Copy
3. Install dependencies:

   pip install -r requirements.txt
   

   Copy
4. For LLM integration, copy .env.example to .env and add your API keys:

   cp .env.example .env
   # Edit .env with your API key
   

   Copy

Basic Socket Implementation

This is a simple implementation using sockets and JSON for educational purposes.

Starting the Basic Server

python mcp_server.py

Using the Basic Client

python mcp_client.py [host] [port]

Official MCP SDK Implementation (HTTP/SSE)

This implementation uses the official MCP Python SDK with HTTP/SSE transport.

Starting the MCP HTTP/SSE Server

python mcp_http_server.py

The server will start on http://localhost:8000.

Using the MCP HTTP/SSE Client

python mcp_http_client.py [server_url]

If no server URL is provided, it will default to http://localhost:8000.

FastMCP Implementation

The server now uses FastMCP, a higher-level wrapper around the MCP SDK that provides a more concise API for creating MCP servers.

Key Features of FastMCP

• Simplified tool and resource registration with decorators
• Automatic type handling and validation
• Streamlined API for creating MCP servers

Starting the FastMCP Server

python mcp_http_server.py

The server will start on http://localhost:8000 and is fully compatible with the existing clients.

LLM Integration with Claude

This example shows how to use MCP with Claude to create an AI assistant that can use tools.

Running the MCP LLM Client

python mcp_llm_client.py [server_url]

If no server URL is provided, it will default to http://localhost:8000.

Available Tools

The MCP server provides the following tools:

1. greet - Greets a person by name
2. calculate - Performs basic calculations (add, subtract, multiply, divide)

Protocol

The official MCP protocol is used in the SDK implementation. For more information, visit modelcontextprotocol.io.

Transport Layers

This project demonstrates different transport layers for MCP:

1. Socket-based - A custom implementation using TCP sockets (for learning)
2. HTTP/SSE - HTTP-based transport with Server-Sent Events for network communication

Extending the Implementation

You can extend the MCP server by:

1. Adding more tools to the server implementations
2. Implementing additional resources
3. Creating specialized servers for specific use cases
4. Implementing other transport layers (WebSockets, gRPC, etc.)

Troubleshooting

导入错误

如果遇到 ImportError: cannot import name 'http_client' 或类似错误，可能是 MCP 库的版本问题。在新版本的 MCP 中，http_client 可能已经被替换为 sse_client。本项目已经添加了兼容性代码，尝试多种导入路径。如果仍然遇到问题，请尝试：

pip install --upgrade mcp

API 密钥问题

使用 LLM 客户端时，确保已经在 .env 文件中设置了正确的 Anthropic API 密钥：

ANTHROPIC_API_KEY=your_api_key_here

Resources

• MCP Official Documentation
• MCP GitHub Repository
• Anthropic Claude Documentation

License

This project is open source and available under the MIT License.