Hi PH! Maker here. I build and debug MCP servers, and I kept hitting the same wall: call a tool through an AI agent and you get the model's paraphrase back, not the real request and response. Wire up a few servers and you've no idea where your tokens are going. So I built MCP Peek — a desktop app to work with MCP servers directly: - Inspect any server (stdio or HTTP/SSE) — every tool, resource & prompt, each with a token-cost estimate - Run tools through a form auto-built from the schema, and read results as table / JSON / Markdown / HTML / image - Proxy — point Claude or Cursor at MCP Peek and watch every call your agent makes, live: timing, tokens, full request/response, plus a topology view It's local-first: your configs, traffic, and secrets stay on your machine (keys in the OS keychain), and the UI makes no network calls of its own. Free on one server with the full inspector; Pro is a one-time $19 (no subscription) for unlimited servers + the proxy. Download's a plain file, no signup. Would love your feedback. What would make this a daily tool for you?

MCP Peek - See what your AI agent really sends to MCP servers

Replies