MCP Proxy Lets Cursor See and Use Thousands of Tools

Discussions
1

Hi everyone,

I’m sharing an early preview of Smart MCP Proxy—a small gateway that lets Cursor IDE tap into hundreds, even thousands, of tools instead of today’s 40-tool cap.

Repo :glowing_star:: GitHub - Dumbris/mcpproxy

Quick start

  1. Install the proxy:
pip install smart-mcp-proxy
  1. Add it as an MCP server in Cursor (see the README for the exact JSON snippet).

I’d love for you to try it out, break it, and tell me what could be better. Bug reports, ideas, and PRs are very welcome.

Thanks for giving Smart MCP Proxy a spin!

3 Likes
2

Love those ideas that enable us to use Cursor in a more effective way.

1 Like
3

@algis can you tell us if your approach matches the research paper at [2505.03275] RAG-MCP: Mitigating Prompt Bloat in LLM Tool Selection via Retrieval-Augmented Generation ?

4

@normalnormie Yes, my proxy implements the same “retrieve-only-relevant-tools” strategy described in the RAG-MCP paper. While the paper relies solely on vector embeddings, I default to BM25 - fast and GPU-free - and let users toggle a configuration flag to enable semantic search. You’ll still see the same reductions in prompt tokens and the selection-accuracy gains reported in the paper.

1 Like
5

Still hacking on this idea - just rewrote mcpproxy in Go.
It’s now faster, has a tray UI, and comes with built-in security features.

If you like the idea of an MCP proxy, give it a try: :glowing_star: ( GitHub - smart-mcp-proxy/mcpproxy-go: Smart proxy that lets your AI agents (Cursor, Claude, ...) use 1000+ tools instead of ~40. It runs offline on your PC with a neat tray UI and even auto-sandboxes new tool plugins for safety )

1 Like