Explore MCP Services with Natural Language

MCP Compass is a discovery and recommendation service for the Model Context Protocol ecosystem itself: it acts as a smart guide that helps an AI assistant find and understand available MCP servers based on natural-language queries, making it easier to discover and pick the right tool for a specific task instead of already knowing what exists.

Use it when you or an AI assistant need to find an appropriate MCP server for a task but don't already know which one to use - describe what you need in plain language and get back relevant MCP services with metadata to evaluate. It is a meta-discovery layer over the MCP ecosystem, not a tool that performs any task itself, so it fits the "which server should I use" question, not doing the actual work once you've found the right one. A hosted version is also available directly at mcphub.io if you want to explore without installing anything.

Smart search finds MCP services using natural-language queries rather than requiring you to already know a server's name. Rich metadata gives detailed information about each discovered service so you can judge whether it fits your task. Results stay current with real-time updates reflecting the latest MCP services as the ecosystem grows, and the service is built for easy integration with any MCP-compatible AI assistant.

Run it directly with npx, or clone and build from source. In Claude Desktop's claude_desktop_config.json, add it as an MCP server:

{
  "mcpServers": {
    "mcp-compass": {
      "command": "npx",
      "args": ["-y", "@liuyoshio/mcp-compass"]
    }
  }
}

Or, if running from a local build:

{
  "mcpServers": {
    "mcp-compass": {
      "command": "node",
      "args": ["/path/to/repo/build/index.js"]
    }
  }
}

Anyone building or using AI assistants who wants a way to discover the right MCP server for a task via natural language, instead of manually browsing MCP server registries. It is released under the MIT License.