User feedback your coding agent can act on.
A feedback button for your app, an inbox your agent reads over MCP, and replies that tell people what you fixed. Built for apps shipped with Claude Code, Cursor and Codex.
Open your inbox- 1
Collect with context
One script tag. Each message arrives with the page, device, release, recent errors and recent clicks, so “it's broken” is something an agent can work from.
- 2
Your agent works the inbox
Connect Claude Code (or any MCP client). It triages, groups duplicates, claims an item, fixes it, and records what changed.
- 3
Close the loop
The person who wrote in hears back, in your app or by email. Feature requests, declines and anything suspicious wait for you.
Safe by default
Feedback is written by strangers and read by an agent that can change your code. Every message reaches the agent marked as untrusted data, never as instructions. Messages that try to talk to an AI, or carry links, commands or secrets, are flagged, and whatever an agent does about them waits for your approval. Reporters' email addresses are never shown to agents.