After testing 340+ MCP servers, one flaw keeps showing up: you can't limit tool exposure
abhiramreddykalluru · reddit · 2026-10-26
After testing 340+ MCP servers, the author of open-source QAmcp found a recurring problem: installing a server for one tool drags in a dozen more, with no way to restrict them. Extra tools are both a security surface (anything can send/modify/delete) and a context clutter (Cursor only uses 40 tools total).
They propose a free open-source proxy that sits between the AI app and MCP servers: allow only chosen tools, default-allow read-only ones while blocking destructive actions until approved, remember approvals, and re-block any tool whose definition silently changes (new description, removed read-only label, hidden instructions). Not built yet — the post asks for feedback on the design.
More from coding & agent
- Suno-MCP: open-source MCP server lets AI assistants generate music via Suno API — modelcontextprotocol · 2026-10-26
- LangChain explains how it builds targeted evals for Deep Agents — sharpeye_wnl · 2026-10-26
- Pointer: open-source MCP server lets AI agents drive your real Chrome tabs — FinancialManager7861 · 2026-10-26
- Open-source Chowbot MCP server lets you order food via ChatGPT or Claude — saheedniyi_02 · 2026-10-26
- What if all human-written code goes in .gitignore? Redditors debate dev's future role — michal-rd · 2026-10-26
- How to run MoEdernColBERT with Sentence Transformers MultiVectorEncoder — tomaarsen · 2026-10-26