Your agent built the app in an afternoon. Your first users found the bugs in ten minutes. Now each report arrives as "it doesn't work" in a DM, and you are the one who has to turn that into something your agent can act on.
The gap: you can't watch every user
When you build with Lovable, Bolt, v0 or Cursor, you rarely read all the generated code. That makes the evidence from the failing session more valuable, not less: which request returned 422, what the console said, what the user clicked just before.
How it works
- Add one script tag (or the npm package) to your app. It works on localhost, preview deployments and production.
- Your user clicks the report button and describes what went wrong. No account needed.
- The report lands with the evidence attached. The element they pointed at, console errors, the failing request with its body and the steps before it.
- Your agent picks it up over MCP. Tell Cursor, Claude Code or Codex to fix the report by its key; it loads the context, proposes the change and resolves the report.
Why not just paste a screenshot into the chat?
A screenshot shows the symptom. The cause is usually in the request body or the error stack, which a screenshot doesn't contain. Reports captured in the user's own browser include both, and the MCP server is on every plan, including Free.