open-multiple-model-mcp-client

# Open Multiple Model MCP Client A flexible client for interacting with multiple Model Control Protocol (MCP) servers through a unified interface, enabling Claude to use tools from various MCP servers. ## Overview This project is designed to serve as a gateway between Anthropic's Claude API and multiple downstream MCP servers. It allows Claude to: - Access tools from multiple MCP servers in a unified way - Invoke tools and receive their results - Compose responses that incorporate tool output ## Features - **Multiple MCP Server Support**: Connect to multiple MCP servers simultaneously - **Flexible Connection Types**: Support for both STDIO and SSE connections - **Dynamic Tool Management**: Enable/disable servers and tools at runtime - **Gateway API**: Expose MCP-compatible endpoints through FastAPI - **Unified Chat Interface**: Simple API for interacting with Claude and tools ## Prerequisites - Python 3.12+ - Anthropic API key ## Installation 1. Clone the repository: ``` git clone https://github.com/yourusername/open-mutiple-model-mcp-client.git cd open-mutiple-model-mcp-client ``` 2. Create a virtual environment and install dependencies: ``` python -m venv venv source venv/bin/activate # On Windows: venv\Scripts\activate uv sync ``` 3. Create an environment file: ``` cp .env-example .env ``` 4. Add your Anthropic API key to the `.env` file: ``` ANTHROPIC_API_KEY=sk-ant-xxxxx ``` ## Configuration ### Environment Variables - `ANTHROPIC_API_KEY`: Your Anthropic API key - `HOST`: Host to bind the server to (default: 0.0.0.0) - `PORT`: Port to run the server on (default: 8000) - `MCP_COMPOSER_PROXY_URL`: URL for the MCP composer proxy (default: http://localhost:8000) - `MCP_SERVERS_CONFIG_PATH`: Path to the MCP servers configuration file (default: ./mcp_servers.json) ### MCP Servers Configuration Configure MCP servers in the `mcp_servers.json` file: ```json { "mcpServers": { "time": { "command": "docker", "args": ["run", "-i", "--rm", "mcp/time"] }, "remote_server": { "url": "https://your-remote-mcp-server.com/mcp" } } } ``` ## Usage ### Starting the Server ``` python main.py ``` By default, the server runs on port 3333 and can be accessed at `http://localhost:3333`. ### Using the Chat API Send a POST request to the `/chat` endpoint: ```bash curl -X POST "http://localhost:3333/chat" \ -H "Content-Type: application/json" \ -d '{"message": "What time is it now?"}' ``` ## API Endpoints - **POST /chat**: Main chat endpoint - Accepts: `{ "message": "your message" }` - Returns: Text response from Claude, including tool use results - **MCP Gateway**: `/mcp/{server_kit_name}` - `/mcp/{server_kit_name}/sse`: SSE endpoint for MCP clients - `/mcp/{server_kit_name}/messages`: HTTP POST endpoint for MCP messages ## Architecture The application follows a layered architecture: 1. **API Layer** (`main.py`): FastAPI application that exposes chat endpoints 2. **Client Layer** (`mcp_client.py`): Handles communication with Anthropic's API 3. **Controller Layer** (`downstream_controller.py`): Manages downstream MCP servers 4. **Composer Layer** (`composer.py`): Orchestrates server kits and gateways 5. **Gateway Layer** (`gateway.py`): Provides MCP-compatible endpoints ## Extending the Client ### Adding New MCP Servers To add a new MCP server: 1. Update the `mcp_servers.json` file with your server configuration 2. Restart the application ### Custom Server Implementations You can implement custom MCP servers by: 1. Creating a new server that follows the MCP specification 2. Adding it to the `mcp_servers.json` configuration