Give an agent email access without pasting secrets in chat
When a built-in connector only covers one Google account, the temptation is to paste an app password into chat for work mail, iCloud, or a custom domain. Inbox Triage avoids that: you open a link yourself, then the agent lists, reads, and drafts over MCP, and sends only after you approve in chat.
A link you open · Encrypted credentials · Send only after you approve in chat
Three workarounds that fail
- Forwarding to a Gmail the agent already has: Sent, drafts, and From stay wrong.
- A new agent-only address: people still email your real address.
- Pasting a password into chat: the secret lands in a transcript and model context.
How Inbox Triage does it
The agent calls a connect tool and gets a URL (Google/Microsoft sign-in or a masked password form). You open it. The credential goes into an encrypted vault (AES-256-GCM). The agent keeps a separate API key for https://getinboxtriage.com/mcp.
What the agent can and cannot do
Can: list inboxes and messages, read bodies (returned as untrusted data), create drafts.
Cannot: see OAuth tokens or IMAP passwords, send mail without your approval in chat, or pull full mailbox history. Sync caps at ~300 newest messages per selected folder for iCloud/IMAP/Microsoft (~40 for Gmail).
Revoke and repair
Disconnect from Account anytime; the stored credential is deleted. To repair a broken connection, open the same connect link again.
FAQ
Is it ever okay to paste a password into chat?
No. Inbox Triage's tools are built so there is never a reason to.
What can the agent see if I don't connect anything?
No real mail: an inbox with no stored credential returns demo data, labeled as demo.
How do I undo a connection?
Disconnect on the Account page. The credential is deleted, not just hidden.
Open the link yourself instead of pasting a password in chat
Encrypted credentials and approve-in-chat send beat a password in a transcript.