Home / Shared inbox for humans and AI agents
Shared inbox for humans & AI agentsMektup gives you a real email inbox that a person and an AI agent operate together. The agent reads, triages, and replies over the API or MCP; you and your team work the exact same mailbox in webmail or in Outlook over IMAP. Let the agent handle routine mail and take over any thread yourself. It's like AgentMail, but humans can use the same inbox too - not a headless, API-only mailbox for a bot.
Today you usually have to choose. An agent-first tool gives your AI a disposable, API-only inbox that no person can open. A human-first shared-inbox tool (Front, Missive, Help Scout) gives your team a nice interface but no real, native way for an autonomous agent to own a thread. So the agent and the humans end up in two different places, looking at two different copies of the same conversation.
Mektup collapses that into one real mailbox with two kinds of operator: software and people, on the same messages, at the same time.
The AI agent connects over the REST API and the native MCP server - it reads, triages, drafts, sends, and replies. You and your team connect to the same mailbox over standard IMAP/SMTP (Outlook, Apple Mail, Thunderbird, your phone) and through Mektup's full webmail dashboard. There's one thread history, and both sides see it.
// the agent, notified by webhook, handles a routine reply list_messages({ mailbox: "support@yourco.com", unread: true }) send_reply({ inReplyTo: msg.id, text: "Thanks! Your reset link is on its way.", }) // meanwhile you're in the SAME inbox in Outlook / webmail: // - you see the agent's reply in the thread // - you grab the tricky ticket and answer it as a human
Because it's one shared inbox, the agent can auto-handle the routine load - triage, acknowledgements, FAQ answers, status replies - while anything it shouldn't touch simply stays in the inbox for a person. You jump into any thread from webmail or your mail client, see the full history the agent saw, and reply as the human. No handoff tooling to build, no context lost between systems.
Every mailbox supports a webhook that fires an HMAC-signed POST the instant new mail arrives, with sender, subject, and body already in the payload - so the agent reacts immediately instead of polling. It's a real inbound + outbound loop on your own domain, not a send-only relay.
| Capability | Mektup | AgentMail | Front / Missive |
|---|---|---|---|
| Real inbox on your own domain | Yes | API inbox | Yes |
| Autonomous AI agent access (API + MCP) | Yes | Yes | No / add-on |
| Human works the same mailbox (IMAP/webmail) | Yes | No | Yes |
| Webhook on new mail, no polling | Yes | Yes | Limited |
| Send transactional email via API | Yes | Yes | No |
| Team logins | Yes | No | Yes |
| Starts at | Free | Paid | Per seat / mo |
If your mental model is "AgentMail, but a person can read and join the same inbox" - or "a shared team inbox that an AI agent can actually run" - that's exactly the middle Mektup sits in. See the deeper Mektup vs AgentMail comparison, or the email for AI agents guide.
Yes - it's the core idea. One Mektup mailbox is a real inbox both a person and an AI agent operate at once. The agent reads, triages, and replies over the REST API or MCP; you and your team work the same mailbox in Mektup's webmail or in Outlook and Apple Mail over IMAP/SMTP. Both act on the same threads in real time.
Mektup. AgentMail gives an AI agent a disposable, API-only inbox with no interface for a person to open it. Mektup gives you a real business mailbox on your own domain that both a human and an agent use at once - the agent over the API and native MCP server, the human over IMAP/SMTP and a full webmail dashboard, with team logins.
Yes. The agent can auto-handle routine mail while anything it shouldn't answer stays in the same inbox for you. Jump into any thread from webmail or your mail client and reply as the human, with full history in front of you.
Every mailbox supports a webhook that fires an HMAC-signed POST the instant new mail arrives, with sender, subject, and body in the payload - so the agent reacts immediately without polling, then reads and replies through the same API or MCP tools.