MCP debugger

Turn an MCP failure into a flow you can read.

When a tool call fails or a model misbehaves, an error message isn't enough. The MCPJam debugger shows every JSON-RPC request and response, the OAuth exchange step by step, and the model's tool choices, so you can find the exact break and reproduce it. Open source, runs locally.

Terminal
npx @mcpjam/inspector@latest

What is an MCP debugger?

An MCP debugger exposes the internals of an MCP interaction so you can find and fix failures. MCPJam shows every JSON-RPC request and response, steps through the OAuth exchange, and runs a real model against your tools so you can see which one it called and why. It reproduces the exact failure locally, an open-source MCP Inspector alternative that goes past protocol validation.

Looking for an MCP Inspector alternative?

The official MCP Inspector is the reference tool for protocol inspection: connect a server, list its tools, and confirm it responds. MCPJam covers that same job and continues past it: a JSON-RPC trace of every message, a visual OAuth debugger, a Playground that runs a real model against your tools, and cross-client testing. Both are free and open source; the difference is how far the debugging goes once a real model is in the loop.

If you're weighing them head to head, the honest breakdown lives on the comparison page.

What you can see and fix

JSON-RPC trace

Read every request and response between client and server in order, so a silent failure becomes a message you can point at.

OAuth Debugger

Step through each phase of the auth exchange between client, MCP server, and authorization server to find where a 401 breaks the flow.

Model-in-the-loop Playground

Run a real model against your server and watch which tool it selects, with what arguments, and how it recovers from errors.

Reproduce any failure

Replay the exact sequence that broke (the same request, the same tool call) instead of guessing from a stack trace.

Cross-client testing

Reproduce a bug in the client it actually happens in (ChatGPT, Claude, Gemini, or Cursor), not just the one on your desk.

Local and open source

No subscription and no data leaving your machine. Point it at any local or remote server and start debugging.

Open source, used in the open

Frequently asked questions

A tool that shows the internals of an MCP interaction (the JSON-RPC messages, the OAuth exchange, and how a model uses the server's tools) so you can find the exact point a failure happens and reproduce it.

Yes. MCPJam Inspector is free and open source, covers the same protocol inspection as the official MCP Inspector, and adds a JSON-RPC trace, an OAuth debugger, a model-in-the-loop Playground, and cross-client testing.

Use the OAuth Debugger to step through each phase of the handshake between client, MCP server, and authorization server, and see the exact request where it breaks. See MCP OAuth testing.

Yes. Point MCPJam at any local or remote MCP server and trace its messages, auth, and tool behavior. A built-in MCPJam tunnel exposes a local server to a real client when you need it.

Stop guessing at MCP failures.

Trace every message, step through OAuth, and reproduce the exact break. Open source, running locally.

Open source · no credit card · ~2 min

Keep exploring