Debugowanie serwera, który nie chce się uruchomić

Sometimes you restart the host and your tools simply aren’t there. Don’t panic: startup failures follow a few predictable patterns. 🛠️

Your first move is always the logs. Claude Desktop writes per-server log files that capture exactly why a process refused to start.

The top cause is a bad command or a relative script path. The host can’t find python or your file, so the process never even runs.

Run the exact command from your config by hand. If it crashes in your terminal, it will crash for the host too, and the error is right there.

An ImportError means the host launched a Python without your packages. Pin the interpreter or use uv so the right environment is used.

A sneaky one: any print to stdout corrupts the handshake, so the client gives up. Move every print over to stderr and try again.

A single trailing comma makes claude_desktop_config.json invalid, and the host loads no servers at all. Validate the file when in doubt.

The simplest miss: you edited config but never restarted. Quit Claude completely, not just close the window, then open it fresh.

Still stuck? Launch the server in the Inspector. If it works there but not in the host, the problem is your config, not your code.

Resist guessing. The traceback in your stderr names the exact line and exception. Read it fully before changing anything. 📋

When you do edit, change a single thing, then retest. Tweaking three settings at once hides which fix actually solved the problem.

A server runs fine in the MCP Inspector but never appears in Claude Desktop. What does that point to?

You learned to read logs, test the command directly, fix paths, deps, and stray prints, and isolate with the Inspector. Your server is verified! 🎉