MCPJam checks your server's compatibility with 18 AI clients (and more to come) when you connect it, including tools, widgets and the client features they need.
This is a static check. If you want to test it, hit "Test" to open the Playground with that client's configuration.
Your local eval passes with --host chatgpt.
That tells you how your server behaves when it sees ChatGPT's handshake. It doesn't tell you ChatGPT can connect to it.
Check out what the emulation covers, and what it doesn't here:
A tool error isn't the end of the turn. The agent decides what happens next: retry with a fix, ask the user, or answer as if the call worked.
The Trace view in the MCPJam Playground shows which one it picked.
docs.mcpjam.com/inspector/play…
Evals and Swarms catch different problems.
An eval scores specific test cases against your MCP server's tools. A swarm sends AI personas on multi-turn journeys and shows where your server fails a real user's goal.
Read more about swarms: mcpjam.com/features/swarms
