How does an AI agent get its own email inbox?

Point the agent at Mektup's MCP server or REST API. It can register a domain, create a mailbox, send mail, list and read incoming mail, and set a webhook so it's notified the instant new mail arrives - all as native tool calls inside its own session.

Can the agent receive email and react to replies, not just send?

Yes. That's the difference from a sending-only API. A Mektup mailbox is a real inbox. The agent reads incoming mail through the API and gets an HMAC-signed webhook on every new message, so it can act on replies and inbound mail without polling.

Can a human read the agent's inbox too?

Yes. Unlike an API-only mailbox, a Mektup inbox is a real one - open it in Mektup's webmail, or add it to Outlook or Apple Mail over IMAP/SMTP - so you can see and manage exactly what your agent is sending and receiving.

Does the agent need its own domain?

No. Every account includes a free, instant inbox on a Mektup subdomain to start immediately. When you want a branded address, the agent (or you) can register any domain you own and Mektup hands back the exact DNS records to add.

A Real Email Inbox for Your AI Agent

Identity, communication, and memory - as a real mailbox

An agent needs an email address for three reasons: identity (to sign up for and verify with the services it uses), communication (to send, receive, and hold threaded conversations), and memory (a durable, searchable record of what was said). Mektup delivers all three as an actual mailbox on real mail infrastructure - not a write-only endpoint - so incoming confirmations, replies, and threads genuinely land somewhere the agent can read and act on.

The full loop, over MCP

Mektup ships a native MCP server, so any MCP-capable agent (Claude Code, Cursor, Lovable, Replit, or your own) drives the whole cycle as tool calls - no glue code, no API keys pasted into generated source:

// point your agent at the Mektup MCP server
{
  "mcpServers": {
    "mektup": {
      "command": "npx",
      "args": ["-y", "mektup-mcp"],
      "env": { "MEKTUP_API_KEY": "mek_live_..." }
    }
  }
}

// then the agent just calls tools, in its own session:
create_mailbox({ domain: "acme.com", localPart: "agent" })
set_mailbox_webhook({ address: "agent@acme.com", url: "https://..." })
send_email({ from: "agent@acme.com", to: "...", subject: "..." })
list_messages({ mailbox: "agent@acme.com" })

New mail fires an HMAC-signed webhook the instant it arrives, so the agent reacts to replies and inbound messages without polling. Every tool maps one-to-one to a documented REST endpoint, so agents that don't speak MCP get the exact same capabilities over plain HTTP.

What makes it a real inbox

Send and receive - real inbound mail, not a sending-only API with no address behind it.
A human can read it too - open the mailbox in Mektup's webmail, or in Outlook or Apple Mail over IMAP/SMTP, and see exactly what the agent sends and receives.
On a domain you own - branded addresses like agent@yourcompany.com, with real DKIM, SPF, and DMARC generated for you.
Delivery status you can trust - sent / deferred / bounced comes from Mektup's own outgoing mail server's real SMTP response, not a tracking pixel.

See it working: a full agent in ~150 lines

There's a complete, MIT-licensed example on GitHub - mektup-inbox-agent - an autonomous email agent that receives mail via the webhook, decides whether to reply or leave it for a human, and answers through the API from the same shared mailbox. Clone it, fill in four env vars, and it runs.

What it costs

Getting started is free - 1 mailbox, 3,000 emails a month, 300 a day, and webhooks included, no credit card. See the full pricing, or how Mektup stacks up in Mektup vs AgentMail.

Set up the MCP server

A real email inbox for your AI agent