Command Mail

What the Command Mail MCP server does

The Command Mail MCP server gives an AI agent access to a unified Command Mail inbox containing one or more email addresses. Depending on the permissions granted, the agent can read and search messages, apply sorting rules, manage folders, prepare drafts, track follow-ups, and send email. Sending is disabled by default until you approve it or grant a broader sending permission.

The repository is an agent-host plugin rather than a local mail processor. It points compatible hosts to Command Mail’s hosted MCP endpoint and includes four skills: inbox-setup, inbox-routine, inbox-cleanup, and mail-workflow. These skills cover account preferences, recurring inbox passes, large-inbox cleanup, and message workflows such as drafting replies and filing conversations.

How it works

The Command Mail MCP server uses Streamable HTTP at https://commandmail.app/mcp. On first use, the agent opens Command Mail in a browser for sign-in. OAuth 2.1 with PKCE supplies the connection token, so the plugin does not ask for an email password or API key.

Access is controlled through permissions and a separate autonomy level for each mailbox. mail:read covers reading, searching, sorting, folders, rules, briefs, and follow-ups. mail:draft permits creating and revising drafts but not sending. mail:send-approved sends the exact revision you approved, while mail:send permits sending under recipient and daily limits. The mailboxes permission covers connecting addresses, domain verification, and importing older mail.

Email content is treated as data rather than agent instructions. You can disconnect an agent from Command Mail settings.

Setup and configuration

Create a Command Mail account before connecting a client. The free account includes one address; Pro adds sending, additional addresses, and full history according to the supplied README.

Claude Code, Cursor, Gemini CLI, and Codex have repository-specific installation paths. Other MCP clients can add the remote server using Streamable HTTP and the endpoint URL. The repository contains host configuration files, skills, and assets, but no executable code, hooks, or installation scripts.

After installation, sign in through the browser and enable only the permissions needed for the mailbox. A recurring routine runs on a schedule only when the agent host supports scheduled tasks and you configure one. No environment variables are required by the plugin.

Tools and capabilities

Supported workflows include:

• Set inbox priorities, autonomy preferences, archive folders, sorting rules, retention settings, signatures, and recurring routines.
• Process new mail since the last completed routine, prepare drafts, record the run, and produce a short brief.
• Review a large inbox in agreed passes, correct sorting decisions, move messages, and create rules.
• Find and read messages, create plain or formatted replies, include attachments, file conversations, and track follow-ups.
• Connect mailboxes, verify domains, and import older mail when the mailbox permission allows it.

The available behavior depends on the Command Mail account plan, mailbox autonomy, and permissions granted during sign-in.

Limitations and notes

The Command Mail MCP server connects to the hosted Command Mail service; the repository does not contain a local email backend. The agent host must support the relevant MCP transport, and scheduled routines additionally require scheduling support from that host.

Sending remains approval-controlled unless you explicitly grant an appropriate permission. Approved sending sends the exact draft revision you approved. Unapproved sending is subject to the service’s stated limit of up to 20 recipients per message and a daily limit.

The plugin communicates with commandmail.app for the MCP endpoint and the OAuth endpoints advertised through its well-known metadata. Privacy, terms, legal, and support information are provided by Command Mail and FYN Labs LLC. The plugin is released under the MIT License.