Skip to main content
Get your AI agent its own email address in two commands.

Prerequisites

1

Install the skill

Inside Claude Code, run:
Claude Code now knows how to use OpenMail.
2

Run setup

In your terminal:
This opens your browser to sign in, prompts for a mailbox name, creates your inbox, and writes credentials to ~/.claude/openmail.env.
3

Verify

Your agent can now send and receive email. Ask Claude Code to send a test email to confirm.

Narrow the agent’s key

Setup signs you in with your account-wide key and stores it for the agent to use. That key reaches every inbox on your account. Once setup has created the agent’s inbox, swap it for an inbox-scoped key — the agent keeps working, but a compromised agent can no longer reach anything but its own inbox.
Replace OPENMAIL_API_KEY in the agent’s env file with the returned token. It is shown once, at creation. Minting needs your account-wide key, so run this yourself rather than handing the account key to the agent.
Setup itself needs the account-wide key — creating an inbox is outside what an inbox-scoped key can do, and the inbox has to exist before a key can be scoped to it. Narrow the key after setup, not before.

More commands

Next steps

WebSockets

Real-time event streaming. No public URL needed.

Email deliverability

Warm-up schedules and content best practices.

Threading

How threads work across inbound and outbound.

API integration

Create inboxes programmatically for multi-tenant apps.