mcp-research-router

MCP Research Router acts as an aggregator and intelligent router for multiple MCP servers. It lets you connect once and access all integrated MCP tools, while automatically recommending the most suitable tool based on your request. It also supports batch parallel execution, enabling faster cross-tool results when you need to combine outputs. You can connect via MCP-compatible clients by loading a small configuration that points at the router, and you can enable an LLM-assisted workflow to get tool recommendations tailored to your query.

Typical usage is via the MCP client configuration, which can be provided through environment variables or a JSON config. At minimum, you connect to the router by spawning it with npx mcp-research-router (or via your orchestrator) and pointing the client to the router’s URL. When configured with LLM capabilities, you can request a list of available tools, ask for intelligent recommendations, and then batch-execute multiple tools in parallel to speed up complex tasks.

Prerequisites:

• Node.js (recommended latest LTS) and npm/yarn installed
• Internet access to install packages

Install and run (one of the following):

1. Global install via npm:

npm install -g mcp-research-router
# then run
mcp-research-router

2. Quick start using npx (no global install):

npx mcp-research-router

Configure the MCP client to connect to the router (see mcp_config example below). If you need to enable LLM support or point at custom servers, set the appropriate environment variables as described in the configuration notes.

Note: If you are integrating into a larger deployment, you can containerize the router or run it as a service behind a reverse proxy as needed.

Tips and notes:

• Recommended environment variables (for enabling LLM and connecting to a server): MCP_LLM_ENABLED, MCP_LLM_BASE_URL, MCP_LLM_API_KEY, MCP_LLM_MODEL, MCP_SERVER_ENABLED, MCP_SERVER_URL, MCP_SERVER_NAME, MCP_SERVER_HEADERS (JSON string).
• MCP_SERVER_HEADERS must be a valid JSON string, e.g. {"Authorization": "Bearer token"}.
• When specifying MCP_SERVER_URL, provide the endpoint where the router exposes MCP tools for your group.
• If you encounter issues with infinite nesting (adding the router to its own group), avoid configuring the router as part of the same MCP group in MCPHub.
• You can still use the router without LLM features; you will have access to tool lists and tool execution, but tool recommendation will be disabled.
• For deployment, consider using a process manager and a reverse proxy to expose the router securely to your clients.