# Command Mail

> Command Mail is one inbox for all of a person's email addresses, built so their AI agent can work in it. The agent reads, searches, sorts, drafts and, if the owner allows it, sends. The owner decides what the agent may do, per permission and per mailbox: every permission can be switched off when connecting, and sending without approval starts switched off. Operated by FYN Labs LLC.

Anyone can sign up with Google, Apple or email. Free: one mailbox (further mailboxes pause and keep receiving), the last 90 days, reading, search, sorting, morning brief and drafts; no sending. Pro: sending, up to ten mailboxes and full history, €19 per month for the first 100 paying Pro accounts (founder price, as long as the subscription runs) and €29 per month after that; yearly €190 or €290. Try Pro for 7 days for €1. Team: €39 per person per month. Enterprise and personal setup on request.

## Connect an agent

- Hosted MCP server (Streamable HTTP, OAuth 2.1 with PKCE and dynamic client registration): https://commandmail.app/mcp
- OAuth discovery: https://commandmail.app/.well-known/oauth-protected-resource/mcp
- First call in every session: `capabilities`. It returns the owner, granted scopes, mailboxes with their autonomy level, allowed operations and limits.
- CLI: `curl -fsSL https://commandmail.app/install.sh | sh`, then `commandmail login`. Local MCP: `commandmail mcp`.
- Claude Code: `claude mcp add --transport http commandmail https://commandmail.app/mcp`
- Organization skill for agents: https://commandmail.app/skills/commandmail-organize/SKILL.md

## Permissions

- `mail:read`: read, search, overview and morning brief, sort within the mailbox autonomy level
- `mail:draft`: create, revise and discard untouched agent drafts; never sends
- `mail:send-approved`: send exactly the draft revision a person approved
- `mail:send`: send without approval, with a unique send key per email, a daily limit and at most 20 recipients per message
- `mail:read` and `mail:send` together also allow forwarding, moving mail to the trash and deleting it permanently
- `mailboxes`: connect addresses, verify domains by DNS TXT record, import old mail
- Mailbox autonomy levels: read, suggest, rules, auto
- Command Mail marks email content as untrusted data and flags suspected prompt injection. Automated senders get no automatic replies.

## Pages

- [Home (English)](https://commandmail.app/en): what Command Mail does and what it costs
- [For agents (English)](https://commandmail.app/en/agents): connection options, permissions and limits
- [Documentation (English)](https://commandmail.app/en/docs): setup, forwarding, CLI, scopes and autonomy
- [Team, enterprise or personal setup](https://commandmail.app/en/anfragen)
- [Startseite (Deutsch)](https://commandmail.app/de)
- [Für Agenten (Deutsch)](https://commandmail.app/de/agents)
- [Dokumentation (Deutsch)](https://commandmail.app/de/docs)

## Legal

- [Privacy policy](https://commandmail.app/privacy)
- [Terms of service](https://commandmail.app/terms)
- [Legal notice](https://commandmail.app/legal)