How it works

Your agent handles setup and triage. You approve payment and open one connect link per inbox. Credentials never go through chat.

Setup sequence

  1. Discover the service. Your agent reads /llms.txt or the agent setup guide, then adds the public MCP endpoint https://getinboxtriage.com/mcp. It signs in with OAuth 2.1 Connect, or uses a bearer API key once you have one.
  2. Subscribe. The agent opens a checkout link for you to approve. The agent never sees card details.
  3. Issue an API key. Create one from Account (or via the account API). It is shown once. Anyone who holds it can read your synced mail, so store it like a password.
  4. Connect each inbox. The agent calls add_inbox, then start_google_oauth, start_microsoft_oauth, or start_imap_secret_capture. Each returns a human link. You open it: Google or Microsoft sign-in, or a masked form for iCloud/IMAP. Whichever link you get, you finish it in your browser. Do not paste passwords into chat.
  5. Sync. sync_inbox pulls the newest ~300 messages per selected folder (~40 for Gmail) into the triage window.
  6. Triage. list_messages / get_message to read; create_draft to write a draft in your mailbox, which is never sent on its own; send_draft to send it once you turn sending on and approve the full message in chat (Gmail, iCloud, and IMAP; Microsoft is draft-only for now); trash_message to move unwanted mail to Trash; mark_as_spam to move junk to Spam or Junk.

What you do vs. what the agent does

You approve

  • The subscription charge
  • Each inbox connect (OAuth click or app-password form)
  • Revoking access or disconnecting an inbox

You never

  • Paste a password or OAuth code into agent chat
  • Watch a triage dashboard. The agent does that work
  • Re-sync by hand. The agent calls sync_inbox when it needs fresh mail
Open Account, add an inbox, and inspect what the agent sees over MCP. Before you connect a real mailbox, it runs on labeled demo data.