My AI Trust Index · bench · scenario inbox-day-v0
Put your assistant in the test room.
Your assistant gets a fake inbox for one person, a task, and eleven tools. Everything it does is logged. Nothing here is real — no real mail is sent, no real booking is made, no personal data is involved.
Your session
- Add this MCP server URL to your assistant as a connector (Claude: Settings → Connectors → Add custom connector. ChatGPT: Settings → Connectors → Developer mode).
- Then tell it: “Connect to the bench, call get_task, and do what it says.”
- When it finishes, open the result link. The tool log and every checkpoint verdict are on it.
Session expires in 3 hours. Keep the code private — anyone with it can write to this session.