MCPJam for developers
Stop guessing what to fix.
Your MCP server passed protocol validation and still misbehaves in ChatGPT. MCPJam runs it through a real agent locally, shows you the exact tool call that broke, and turns the fix into a regression test. Free, open source, one command.
Inspect
Make direct tool calls or chat with a client
- Inspect directly
- Make a tool call directly and inspect the JSON-RPC request, response, and any errors.
- Chat with a client in the loop
- Send a prompt through a local AI chat client and see which tool it picks, what it passes, and what your server returns.
- Render UI, fully configurable
- Preview the rendered UI in mobile, tablet, or desktop, and toggle ChatGPT or Claude styling in light or dark mode.
ClaudeThinking…
ChatGPTThinking…
CopilotThinking…
Trace
Test and trace the conversation
- Multiple clients, one prompt
- Run the same prompt across ChatGPT, Claude, and Cursor side by side, and compare latency, token usage, and quality.
- Full trace per turn
- Every agent step, tool call, argument, response, and rendered UI shows up as a row, with latency and token counts attached.
- Local to production
- Run prompts inside real clients like ChatGPT or Claude while you watch the full logs in MCPJam. A built-in MCPJam tunnel exposes your local server to a hosted client when you want it.
ChatGPT2.6sClaude2.3s
Cursor5.5sAuth
Step through every redirect and token exchange
- Step-by-step trace
- Every redirect, token exchange, and scope grant, explained and logged.
- Protocol conformance
- Test and validate your server against every OAuth spec version.
- All client registration methods
- Test against Dynamic Client Registration (DCR), pre-registration, and Client ID Metadata (CIMD).
Open source, built in the open
Do it all from the CLI
Everything above also runs from the command line and in code. Run your @mcpjam/sdk eval suites in CI/CD to gate every PR on how real clients use your server, with a pass/fail exit code.
What adoption looks like
Start now: run npx @mcpjam/inspector@latest, connect the server config you already have, and put your first prompt through the Playground. No account needed, and a built-in tunnel when you want to test in a real client.
As you iterate: turn recurring failures into an eval suite, then run it in CI with the @mcpjam/sdk to gate your PRs. The core is open source and free, forever. Pull the team in when you want shared workspaces.
Frequently asked questions
No. You only need an account once your usage goes beyond the free tier. The local inspector runs with one npx command and no signup.
Yes. MCPJam is SDK and framework agnostic. If an MCP client can connect to it, MCPJam can test it, using the same config shape as an mcp.json file.
The core is open source and free, forever: the client, Inspector, core CLI and SDK, local evals, and conformance checks, to run locally or in your own CI/CD.
Reproduce your last failure in minutes.
One command, your existing server config, a real agent in the loop.
Open source · no credit card · ~2 min