The Wireshark of AI
A native desktop inspector for Model Context Protocol JSON-RPC. Live capture, replay, diff, and export across stdio, HTTP, and SSE — with zero changes to your servers or clients.
No data leaves your machine on Free and Pro.
What it does
Built for developers who are tired of adding console.log to their MCP server source and hoping for the best.
Every JSON-RPC message, both directions, as it happens. MCP Inspector handles client proxy config — your server needs zero modification.
Pick any captured request and fire it again against the live server. See the new response side-by-side with the original.
Compare any two traces — params, results, or errors — with a collapsible JSON diff. Catch response shape changes before they hit production.
Save any session as a .mcp-trace file. Attach to a bug report. Open on any machine running MCP Inspector.
Color-coded timeline shows every call status — in-flight, done, slow (≥200ms), or error. Find the bottleneck without reading logs.
Free and Pro run entirely locally. No telemetry, no cloud sync. Team tier: self-hosted aggregator on your own infrastructure.
Pricing
The free tier is a real product, not a demo. Use it indefinitely. Upgrade when you need replay, diff, or team visibility.
Free
Portable .exe. No install, no account. One command captures live MCP traffic.
Pro
Everything in Free, plus unlimited history, replay, diff, and export. Two machine licenses.
.mcp-traceLaunching soon — get notified first
Team
Self-hosted aggregator on your infrastructure. Every developer sees every agent's traffic on a shared dashboard. Min 3 seats.
Enterprise: self-hosted license server, SSO/SAML, audit logs, air-gap mode. Contact us →
FAQ
stdio (subprocess wrap), http (direct streamable-HTTP), and sse (legacy SSE transport). The reverse proxy handles SSE and HTTP; the MITM wrapper handles stdio. You pick based on how your server runs.
mcp-inspector) runs on any platform where Python 3.11+ is available.