MCP Router

MCP Router is a service discovery and proxy for MCP (Model Control Protocol) services. It allows you to register, discover, and execute tools on remote MCP servers.

Features

1. Service Discovery: Register and search for MCP services using vector-based similarity search
2. Tool Execution Proxy: Execute tools on remote MCP servers through the router
3. MCP Compliance: Fully compliant with the MCP protocol specification

Architecture

The MCP Router consists of two core modules:

1. MCP Data Plane: The frontend that interacts with MCP clients, providing service discovery, registration, and tool execution proxying
2. MCP Discovery Service: The backend that uses Alibaba Cloud's vector database and embedding models for intelligent search and management of registered MCP services

Prerequisites

• Python 3.11 or higher
• Access to Alibaba Cloud DashScope and DashVector services
• API keys for DashScope and DashVector

Installation

1. Clone the repository:

   git clone <repository-url>
   cd mcp-router
   

2. Install dependencies using uv:

   uv add pydantic httpx dashscope dashvector python-dotenv
   

3. Set up environment variables in .env:

   DASHSCOPE_API_KEY=your_dashscope_api_key
   DASHVECTOR_API_KEY=your_dashvector_api_key
   DASHVECTOR_ENDPOINT=your_dashvector_endpoint
   COLLECTION_NAME=mcp_services_collection
   EMBEDDING_DIMENSION=1024
   

Usage

Run the MCP Router server:

uv run python mcp_router.py

Tools

The MCP Router provides three tools:

1. search_mcp_server: Search for registered MCP servers based on a query
2. add_mcp_server: Register a new MCP server with the discovery service
3. exec_mcp_tool: Execute a tool on a target MCP server

Development

Running Tests

uv run pytest

Code Structure

• mcp_router.py: Main MCP server implementation (Data Plane)
• discovery_service.py: Discovery service implementation (Discovery Service)
• config.py: Configuration management
• test_mcp_router.py: Unit tests

License

MIT