MCP Server
[ ](README.zh-CN.md)
Project Overview
Built on FastAPI and MCP (Model Context Protocol), this project enables standardized context interaction between AI models and development environments. It enhances the scalability and maintainability of AI applications by simplifying model deployment, providing efficient API endpoints, and ensuring consistency in model input and output, making it easier for developers to integrate and manage AI tasks.
MCP (Model Context Protocol) is a unified protocol for context interaction between AI models and development environments. This project provides a Python-based MCP server implementation that supports basic MCP protocol features, including initialization, sampling, and session management.
Features
- **JSON-RPC 2.0**: Request-response communication based on standard JSON-RPC 2.0 protocol
- **SSE Connection**: Support for Server-Sent Events connections for real-time notifications
- **Modular Design**: Modular architecture for easy extension and customization
- **Asynchronous Processing**: High-performance service using FastAPI and asynchronous IO
- **Complete Client**: Includes a full test client implementation
Project Structure
Installation
- Clone the repository:
- Install dependencies:
Usage
Starting the Server
By default, the server will start on `127.0.0.1:12000`. You can customize the host and port using environment variables:
Running the Client
Run the client in another terminal:
If the server is not running at the default address, you can set an environment variable:
API Endpoints
The server provides the following API endpoints:
- **Root Path** (`/`): Provides server information
- **API Endpoint** (`/api`): Handles JSON-RPC requests
- **SSE Endpoint** (`/sse`): Handles SSE connections
MCP Protocol Implementation
Initialization Flow
- Client connects to the server via SSE
- Server returns the API endpoint URI
- Client sends an initialization request with protocol version and capabilities
- Server responds to the initialization request, returning server capabilities
Sampling Request
Clients can send sampling requests with prompts:
The server will return sampling results:
Closing a Session
Clients can send a shutdown request:
The server will gracefully shut down:
Development Extensions
Adding New Methods
To add new MCP methods, add a handler function to the `MCPServer` class and register it in the `_register_methods` method:
Integrating AI Models
To integrate actual AI models, modify the `handle_sample` method:
Troubleshooting
Common Issues
- **Connection Errors**: Ensure the server is running and the client is using the correct server URL
- **405 Method Not Allowed**: Ensure the client is sending requests to the correct API endpoint
- **SSE Connection Failure**: Check network connections and firewall settings
Logging
Both server and client provide detailed logging. View logs for more information:
References
- [MCP Protocol Specification](https://www.claudemcp.com/specification)
- [FastAPI Documentation](https://fastapi.tiangolo.com/)
- [JSON-RPC 2.0 Specification](https://www.jsonrpc.org/specification)
- [SSE Specification](https://html.spec.whatwg.org/multipage/server-sent-events.html)
License
This project is licensed under the MIT License. See the LICENSE file for details.