mcpwall

A transparent stdio proxy that sits between your AI client and any MCP server, enforcing YAML-defined security policies on every JSON-RPC call. It blocks dangerous tool invocations (SSH key access, `rm -rf`, pipe-to-shell), scans arguments and responses for leaked secrets using regex and entropy analysis, and logs everything to a JSON Lines audit trail. Ships with sensible defaults but supports custom rules for path restrictions, command patterns, and secret detection. Works with Docker MCP Toolkit, Claude Code, Cursor, and any stdio-based MCP server. Think iptables for your AI toolchain: no LLM decisions, no cloud calls, just deterministic rules that stop your assistant from reading `.env` files or exfiltrating API keys before the request hits the filesystem server.