# MCPProxy

A local proxy in front of every MCP server you run: the agent loads a short list of search tools, and new servers stay quarantined until you approve them.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed · ★ 398
- Author: smart-mcp-proxy
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/smart-mcp-proxy/mcpproxy-go
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/mcp/mcpproxy

## What it does

MCPProxy runs as a local process on macOS, Linux and Windows. The core is one binary with the web UI inside it, so there is no extra service to launch; macOS also gets an optional menu-bar app for start, stop, server health and logs. It federates the MCP servers you connect, which lets you go past client caps such as Cursor's 40-tool limit and the 128-function ceiling in OpenAI.

Instead of the full upstream catalog, the agent sees the proxy's own discovery and execution tools — `retrieve_tools`, `describe_tool` and the `call_tool_*` variants — then fetches schemas when it needs them. Context savings vary with tool count, routing mode and client. On the frozen 45-tool reference corpus the initial proxy menu is actually larger than plain listing, since its own tools have a fixed cost; savings only kick in past the break-even point.

Fresh servers are quarantined and their tools withheld until you review them. Under the default `manual` trust mode, edits to an approved tool's definition come back for another review. That narrows exposure to Tool Poisoning Attacks but is no safety guarantee. Docker-based scanners — Snyk, Semgrep, Trivy, Cisco — can inspect quarantined servers before approval, and findings are normalised to SARIF with a composite risk score.

## Requirements
- Needs: a prebuilt binary from the repository releases

## Install

### Claude Code (native platform)

**Install**
- The MCP registry has no launch command for it: see the repository README

## When to use it
- connect many MCP servers without hitting a client tool limit
- vet a new server's tools before the agent can call them
- shrink the tool menu the agent loads at startup
- scan a quarantined server with security tools before approval
- watch MCP activity from the macOS menu-bar app

## Example prompts
- Use `retrieve_tools` to show what the connected servers can do
- Approve the quarantined server once I review its tools
- Describe that tool's schema with `describe_tool` before calling it

## Similar
- [GitHub](https://affpapa.org/en/nevibecode/mcp/github.md) — MCP server · ★ 33918
- [Netdata](https://affpapa.org/en/nevibecode/mcp/netdata.md) — MCP server · ★ 81115
- [browser-use](https://affpapa.org/en/nevibecode/mcp/browser-use.md) — MCP server · ★ 119160
- [Codebase Memory](https://affpapa.org/en/nevibecode/mcp/codebase-memory.md) — MCP server · ★ 47392
- [mcp-server](https://affpapa.org/en/nevibecode/mcp/mcp-server.md) — MCP server · ★ 43872
- [AIHawk](https://affpapa.org/en/nevibecode/mcp/aihawk.md) — MCP server · ★ 32221

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md