Skip to content
Blog / News

AgentMail is now an official OpenClaw plugin

Adi Singh

AgentMail is on ClawHub as an official OpenClaw plugin. It ships a CLI-backed AgentMail skill plus an email channel: inbound mail drives an agent turn and the agent replies in the thread, with a default-deny sender allowlist. Install it with one command.

  • openclaw
  • openclaw plugin
  • official openclaw plugin
  • openclaw email plugin
  • openclaw email channel
  • clawhub
  • agentmail openclaw
  • openclaw email
  • openclaw plugins install
  • announcement
  • integrations
  • agent-infrastructure

TL;DR

  • AgentMail is now an official plugin on ClawHub, the OpenClaw plugin registry.
  • Install it with openclaw plugins install clawhub:@agentmail/agentmail.
  • It ships two things: a CLI-backed AgentMail skill, and an email channel for OpenClaw. Inbound email drives an agent turn, and the agent replies inside the AgentMail thread.
  • The channel is reply-only and default-deny. Unknown senders are rejected, replies stay bound to the message that triggered them, and the agent cannot start threads or mail arbitrary recipients.

Nat Eliason describing his AgentMail setup for OpenClaw, and Peter Steinberger replying that there should be an official plugin and channel

That was five weeks ago. He asked for a plugin and a channel, and both now exist.

AgentMail is published on ClawHub, the OpenClaw plugin registry, with the Official badge.

Bash

openclaw plugins install clawhub:@agentmail/agentmail

The fit is structural rather than lucky. OpenClaw's premise is that you reach your agent wherever you already are, and it goes and does the thing. An agent that acts on your behalf needs two things: somewhere to be reached, and something to be recognised by. Email is both, and it is the only address that every person, company and machine already accepts without being asked to install anything first.

Which makes the gap on the shelf odd. ClawHub lists twenty-five channel plugins for OpenClaw. WhatsApp, Matrix, Discord, Slack, Teams, Feishu, LINE, Google Chat, QQ, Zalo, Nostr, Twitch, Urbit, iMessage through BlueBubbles. Twenty-five ways to reach an agent, and until now not one of them was email.

What is OpenClaw?

OpenClaw is an open source AI agent that runs on your own hardware rather than in someone else's cloud. It reads and writes files, runs shell commands, browses, and keeps long-term memory locally.

The scale is worth stating plainly: 386,000 stars and 81,000 forks on GitHub, from a repository that did not exist before November 2025.

Channels are the part that matters here. OpenClaw does not have a UI so much as it has inboxes. You message it on WhatsApp or Discord or Matrix, it does the work, it messages you back. Every capability it has arrives through a channel, which is why the missing one was worth fixing.

What is AgentMail?

AgentMail is email infrastructure built for agents instead of people. You create an inbox with an API call and get a real, addressable email address in under a second. No OAuth consent screen, no per-seat billing, no human sitting behind the account.

Inboxes are two-way. Your agent sends, receives, and replies inside real threads with the message history intact, and every operation available in the dashboard is also an API endpoint. Incoming mail can fire a webhook the moment it lands.

Why does everyone running OpenClaw need this?

Because until now your agent could reach you, and almost nobody else. Chat channels connect an agent to the people already in that app. Email connects it to everyone, including the systems that only speak email: verification codes, invoices, calendar invites, support queues, and other agents.

Email becomes a channel. Before this, giving OpenClaw an email address meant wiring it to Gmail through OAuth or IMAP and hoping Google did not flag the account. Now email works the way WhatsApp does. Someone emails the agent's address, that message drives an agent turn, and the reply goes back in the same thread.

In practice that means you can put your own address in allowFrom, forward your agent a thread from your phone with "draft me a reply to this," and get the draft back in that thread. No terminal, no session, no laptop open.

The defaults are narrow on purpose. Email is the easiest prompt-injection surface anyone will point an agent at, so the channel is locked down before you touch it:

  • dmPolicy defaults to allowlist, and an empty allowlist denies every sender. You opt senders in, not out.
  • Replies are bound to the message that triggered them. The agent cannot open new threads or mail addresses you never approved.
  • Every reply re-hydrates and re-authorizes the original sender, so a forged Reply-To cannot redirect where the response goes.
  • Inbound mail is committed durably before it is acknowledged, so a restart mid-turn does not lose the message.

The failure mode is "nothing happens," not "a stranger drove my agent."

The skill does not go stale. Instead of a fixed set of hardcoded tool schemas, the plugin bundles the official AgentMail CLI and teaches the agent to read its help output. When AgentMail ships a new resource, your agent can use it without waiting for a plugin release. It already covers inboxes, messages, threads, drafts, webhooks, domains, pods, and API keys.

How to get it

Install and enable the plugin:

Bash

openclaw plugins install clawhub:@agentmail/agentmailopenclaw plugins enable agentmail

Put your credentials where the Gateway can read them, in ~/.openclaw/.env. Grab a free API key from the AgentMail console. The webhook secret is optional; without it the channel falls back to WebSocket ingress.

Code

AGENTMAIL_API_KEY=am_...AGENTMAIL_WEBHOOK_SECRET=whsec_...

Point the channel at an inbox and list who is allowed to talk to it, in your OpenClaw config:

Code

{  channels: {    agentmail: {      apiKey: { source: "env", provider: "agentmail", id: "AGENTMAIL_API_KEY" },      inboxId: "agent@agentmail.to",      webhookSecret: { source: "env", provider: "agentmail", id: "AGENTMAIL_WEBHOOK_SECRET" },      dmPolicy: "allowlist",      allowFrom: ["you@example.com"],    },  },}

Restart the Gateway and confirm it loaded:

Bash

openclaw gateway restartopenclaw plugins inspect agentmail --runtime

The CLI-backed skill is available to the agent immediately, and you can drive it yourself with a passthrough command:

Bash

openclaw agentmail -- --format json inboxes list

Requirements: OpenClaw 2026.7.2 (beta) or newer, and Node 22.22.3+, 24.15.0+, or 25.9.0+. The published plugin bundles the AgentMail CLI for macOS, Linux, and Windows, so there is nothing else to install.

Where to go next

FAQ

Questions, answered.

Install the official AgentMail plugin with openclaw plugins install clawhub:@agentmail/agentmail, then openclaw plugins enable agentmail, and set AGENTMAIL_API_KEY in the environment that runs the Gateway. Your agent can then create inboxes on demand, each with a real address, with no Google account and no OAuth consent screen.

Configure the plugin's email channel under channels.agentmail with an inboxId and an allowFrom list. Inbound mail from an allowlisted sender drives an agent turn and the agent replies inside the same thread. There is no polling loop to write: set AGENTMAIL_WEBHOOK_SECRET for Svix-signed webhook ingress, or leave it unset and the channel falls back to WebSocket.

Give it its own inbox rather than access to yours. AgentMail creates an addressable inbox by API in under a second, so the agent owns an address instead of borrowing your Google account. Nothing is shared, nothing can be revoked out from under it, and a suspension cannot take your own mail down with it.

Email is the most common prompt-injection surface for an agent, so the channel is locked down by default rather than by configuration. dmPolicy defaults to allowlist and an empty allowFrom rejects every sender. Replies stay bound to the message that triggered them, every reply re-authorizes the original sender so a forged Reply-To cannot redirect delivery, and inbound mail is committed durably before it is acknowledged.

Yes. AgentMail is published on ClawHub, the OpenClaw plugin registry, under @agentmail with the Official badge. Install it with openclaw plugins install clawhub:@agentmail/agentmail. ClawHub lists twenty-five channel plugins for OpenClaw and every other one is a chat surface, so AgentMail is the email path.

A skill is a markdown instruction bundle that teaches the agent how to use tools it already has. A plugin is a code package that registers new runtime surfaces: channels, tools, services, HTTP routes. The AgentMail plugin ships both. It carries a CLI-backed AgentMail skill and registers an email channel, which a skill alone cannot do.

Blog

Suggested reading

View all posts

Prompt

Give yourself access to AgentMail by following the guide at:

https://docs.agentmail.to/agent-onboarding.md

and refer to:

https://docs.agentmail.to/llms.txt

for the documentation index.

Start integrating AgentMail into your AI agents today.