its hard to debug my mcp when i can’t export tool calls. be default its just question and answer
It’s probably not relevant anymore after two months, but in my case, I just added a debug option that logged everything in detail in the response — and that was enough for the Agent to troubleshoot the MCP. I already had a verbose flag baked into the design, so maybe the debug mode just reused its output.
My MCP is made in Python and I installed it in editable mode in the venv of the itself, so backend edits didn’t even require me to manually reconnect the MCP to Cursor.