Mnki Mcp

A local MCP proxy that checks each tools/call against Agent Trust policy, with evidence, before forwarding it to the wrapped MCP server.

What you can do

Intercept tools/call requests between an MCP client and a wrapped MCP server, whether that server is a stdio command or a Streamable HTTP endpoint, and verify each call against Agent Trust policy before it runs, returning evidence with the verdict. Forward all other MCP traffic untouched. Run in observe, warn, require_approval or enforce mode, from log-only to denying calls and waiting for human approval. Surface JSON-RPC errors for denied calls, approval required or rejected or expired, verification quota exceeded, and Agent Trust being unreachable.

When to use it

Use it when you want a policy checkpoint in front of an existing MCP server so that tool calls are verified and evidenced before execution, and when you need graduated rollout from logging to enforcement.

Requirements

Runs locally over stdio, typically via npx from the mnki-mcp npm package, so Node.js is needed. Requires an Agent Trust API key with the verify scope in MNKI_API_KEY; MNKI_URL optionally points at a self-hosted console instead of the default. Network access to the Agent Trust service is needed for verification. Desktop clients only.