Overview

AI-generated overview

Connects an AI assistant to a hosted Command Mail inbox so it can read, search, sort, draft and, with permission, send email.

What it does: Command Mail is a hosted email service that an assistant works inside. Through the remote MCP server, the agent can read, search and sort mail, prepare an overview and morning brief, track follow-ups, and create or revise drafts. Sending is gated by permissions: it can send only an approved draft revision, or send without approval if that permission is granted. It can also connect addresses, verify domains and import old mail.
When to use it: Useful when you want an assistant to triage a busy inbox, summarise and sort messages, track follow-ups, and prepare replies across several email addresses. Worth adding if you already use Command Mail or are willing to create an account, and you want per-mailbox control over what the agent may read, draft or send.
Requirements: A remote MCP server reached over Streamable HTTP at the provider's endpoint; no local runtime or package is needed. A Command Mail account is required, and sending, extra addresses and full history are part of the paid Pro tier. Sign-in happens through OAuth 2.1 with PKCE in the browser; no password or API key is entered into the plugin. The agent host talks only to the provider's domain.

Before you install

The agent can act on your email: reading and searching mail, creating drafts, and, if you grant it, sending messages without approval (up to 20 recipients per message and within a daily limit). Sending without approval starts switched off, and each mailbox has an autonomy level limiting changes. Email content is treated as data rather than instructions. Access is granted through an OAuth token, which you can revoke under Settings → Connections.

Installation

In SourceWeft

Open Command Mail in the dashboard and add it to a workspace.
Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "mail": {
      "type": "http",
      "url": "https://commandmail.app/mcp"
    }
  }
}

README

Command Mail plugin

Command Mail is one inbox for all of your email addresses, built so your AI agent can work in it. The agent reads, searches, sorts and drafts, and sends only if you allow it, per permission and per mailbox.

This repository connects agent hosts to the hosted Command Mail MCP server at https://commandmail.app/mcp and adds a mail workflow skill. You need a Command Mail account: start for free with one address; Pro adds sending, more addresses and your full history.

The server is also listed in the Official MCP Registry as app.commandmail/mail.

Install

ChatGPT and Codex. Command Mail will appear in the ChatGPT and Codex plugin directory once it is listed there. Until then, add this repository as a marketplace in Codex:

Claude Code.

Cursor. Clone the repository into your local plugin folder, then run Developer: Reload Window:

Gemini CLI.

Any other MCP client. Add a remote MCP server with the Streamable HTTP transport and the URL https://commandmail.app/mcp.

Sign in

On first use, your agent opens Command Mail in the browser. Sign in and switch on the permissions you want to grant; sending without your approval starts switched off. The connection uses OAuth 2.1 with PKCE, so you never give the agent a password or API key. You can disconnect an agent at any time under Settings → Connections.

Permissions

Permission	What the agent may do
`mail:read`	Read, search and sort mail, prepare the overview and morning brief, track follow-ups
`mail:draft`	Create and revise drafts; never sends
`mail:send-approved`	Send exactly the draft revision you approved
`mail:send`	Send without approval, once per message, to at most 20 recipients and within a daily limit
`mailboxes`	Connect addresses, verify domains, import old mail

Each mailbox also has an autonomy level (read, suggest, rules or auto) that limits what an agent may change. Email content is treated as data, never as instructions.

Network and credentials

The plugin ships no executable code, hooks or install scripts. It declares one remote MCP server, https://commandmail.app/mcp, and the agent host talks only to commandmail.app: the MCP endpoint and the OAuth endpoints it advertises under /.well-known/. The only credential is the OAuth token your agent host receives after you sign in; you never enter a password or API key into the plugin.

Package layout

plugin.json also carries the listing, review cases and release notes for the OpenAI plugin directory under extensions.com.openai. Build the upload ZIP with scripts/build-openai-zip.sh /path/to/commandmail-plugin.zip; it contains only the portable package. Reviewer credentials never belong in this repository.

Legal and support

Command Mail is operated by FYN Labs LLC. This plugin is released under the MIT License.

Source: README.md at commit 736a1d8

Tools

0

Tool metadata has not been indexed yet.

Version history

1

v0.2.0LatestOct 3, 2026

Host	Files
ChatGPT, Codex, Cursor (Agent Plugins)	`plugin.json`, `mcp.json`
Claude Code	`.claude-plugin/`, `.mcp.json`
Gemini CLI	`gemini-extension.json`
All hosts	`skills/mail-workflow/`, `assets/`

Command Mail