Not a send-only API. Your agent gets its own address and the full set of email tools it drives over one MCP server.
Each agent gets a real, sendable mailbox on your subdomain — one it receives to, replies from, and can be reached at, kept apart from your own inbox.
Connect through one MCP server and your agent can read, list, search, reply, forward, and compose — the whole inbox, exposed as tools.
Every key ships a downloadable skill bundle with the MCP connection and the key already baked in. One step to point an agent at its mailbox.
Mint a separate key per agent, each with its own note and permissions — and revoke any one of them the moment you need to, without touching the rest.
The agent proposes; you decide. Nothing leaves your mailbox until it clears your gate.
Replies the agent writes wait in a hold queue for your review — accept or reject each one before anything goes out in your name.
Grant one key trusted auto-send for routine mail and keep another on a tight leash. Different agents, different keys, different permissions.
An advisory guard flags incoming email that tries to manipulate your agent into acting outside its scope, so a poisoned message gets a second look.
See exactly what the agent read and every reply it proposed. Nothing it does happens off the record.
From switching on the mailbox to a working agent — without building the inbox, the tools, or the oversight yourself.
Enable the MCP server for a mailbox and set the scope you want the agent to operate within.
Create a scoped, revocable key with a note, then download its ready-to-use skill bundle.
Your agent reads, searches, and proposes replies over MCP — and you approve what actually goes out.
Collega una casella postale, carica un documento, guarda Mailon redigere la sua prima risposta. Nessuna carta richiesta.