Send agent feedback
Tell the administrators of this instance about a problem with a tool. This is what the submit-feedback MCP tool calls. The person the agent works for never sees the report in Monica, only the call in their MCP history.
Send it at most once per task, and only when a tool got in the way. Paraphrase what the person asked, and use placeholders such as <contact> or <date> instead of names, dates and other details from the vault. Never include secrets.
Body parameters
Sent as JSON. Anything not listed here is ignored.
bug when a tool failed or returned something wrong, missing_tool when no tool could do it, unclear_description when a description misled you, improvement for a suggestion.
blocks_task when the task could not be finished, needs_workaround when it was finished another way, minor when it was finished with friction.
One line naming the problem, up to 200 characters.
The tool the report is about, or the name of the tool that is missing, up to 100 characters.
What the person asked, paraphrased, up to 4,000 characters.
The tool calls tried, in order, at most 20. Each holds tool (required), arguments as a string, outcome (success or error, required) and result.
What happened, up to 4,000 characters. A placeholder such as test, todo or n/a answers 422.
What should have happened, up to 4,000 characters.
The fix you suggest, up to 4,000 characters.
Code for the fix, up to 8,000 characters.
The model the agent runs on, up to 100 characters.
Returns
The id of the report and its status.
The id of the report.
Always new.
Rate limit
Ten requests per minute, per user. Every answer carries the budget left in X-RateLimit-Remaining.
Keys act as you, with exactly the access you have. Make one per integration so you can revoke it alone.