Ncp

If you're running multiple MCP servers and your AI is drowning in tool schemas, this orchestration layer sits between your client and all your MCPs to solve the cognitive overload problem. Instead of exposing 50+ tools directly, it gives your AI just two or three: find (semantic search across all available tools), code (execute TypeScript that calls any underlying tool), and optionally run (individual tool execution). It handles routing, caching, and discovery automatically. The big win is token efficiency: testing shows 97% reduction in schema tokens for large setups (80 tools goes from 103k tokens to 2.5k). Ships with project-level config so each workspace can define its own MCP stack. Best suited for power users juggling filesystem, database, GitHub, Slack, and other MCPs who want their assistant to stop asking which tool to use and just get work done.