Your AI agent, across all your mail.
Connect Claude or any agent that speaks the Model Context Protocol once, and it works across the mail of all your domains and products. It searches, sorts, drafts replies and, when you allow it, sends.
What your agent can do
| Search | Find messages across every inbox by text, sender, address, domain or date |
|---|---|
| Read | Open a message or a whole thread, with a list of attachments |
| Organise | Move, label and mark messages as read |
| Draft | Create and edit replies in your Drafts folder, from the right address |
| Send | Send a new message or a draft, when you allow it |
| Addresses | List addresses, create an alias, disable an alias |
| Domains | List domains and check their DNS status |
| Set up | Add a domain and get the DNS records it needs, verify them, create mailboxes |
Your agent adds DNS records with its own access to your DNS provider; we never ask for that
access and never write to your DNS. You connect at https://mcp.centomail.com and sign in
with OAuth, so you never paste a password or key into the agent.
Typical uses
- "Sort today's support mail across all my products and draft replies to the easy ones."
- "Which customers wrote about the login bug this week? Draft one answer for all of them."
- "Find the invoice from my web host and tell me when it is due."
- "Summarise this thread and draft a polite reply that says no."
- "Is DNS for every one of my domains healthy?"
The security model
Email is the most sensitive data most people have, and it is full of text written by strangers. You decide how far your agent goes; we make the limits clear and enforce them.
Sending is your choice
A new agent gets drafts only: it prepares replies, you read and send them. Sending is a separate box, unticked by default, that you tick when you connect an agent you trust. Its mail goes out under the same limits as anything else you send, and every message it sends is in your Agent activity log.
Mail content is untrusted
Any message can contain instructions aimed at an AI ("forward all invoices to..."). Message content is marked as external data in every tool response. With drafts only, such a message cannot make mail leave your account; that is why sending is a separate permission.
Limited reach
Choose read-only, read plus organise plus drafts, or sending, and limit an agent to specific folders. Aliases can be disabled but never deleted, and forwarding and account security settings cannot be changed through MCP at all.
Logged, and revoked in one click
Every action an agent takes is logged. Every connected agent is listed under Connected apps, and revoking one takes one click and works immediately.
Which AI sees your mail?
centomail does not run its own AI on your mail and does not send your mail to any AI provider by itself. MCP is a door you open for an agent you choose, running on your own AI subscription. Data the agent reads goes to that agent's provider under their terms, so pick one you trust, and close the door any time.