Claude
Web and desktop: connect Command Mail from the Claude directory, where it is listed as a community connector. Or add a custom connector with the URL https://commandmail.app/mcp. Confirm access on the authorization page.
DocsFor people and agents
From your first address to a connected agent. You can start free and set everything up yourself.
https://commandmail.app/mcp to your agent; Claude Code has a one-line command. Instructions by client.Signing in with Google shares only your name, email address, profile picture link and account ID. Command Mail gets no access to Gmail, Drive or Calendar.
Command Mail is one inbox for multiple addresses and domains you own. Email arrives through forwarding, for example from Gmail or your own domain. Each address gets its own private inbound address.
Use the browser or CLI. Then set up forwarding with your existing provider to that address’s private inbound address.
commandmail mailboxes connect mara@beispiel.deVerify ownership of your domain with a DNS TXT record: first claim it, then verify it.
commandmail domains claim studio-beispiel.de
# Add the TXT record at your DNS provider
commandmail domains verify studio-beispiel.deImport MBOX and EML exports or folders of EML files through the CLI. First connect the destination mailbox and log in to the CLI. For example:
commandmail import export.mbox --mailbox mara@beispiel.de --state archiveImport brings in existing messages. It does not set up continuous sync with your previous provider.
If your agent has the scope mailboxes, it can handle these steps. For example, say: “Set up hallo@studio-beispiel.de in Command Mail.” It connects the address, claims the domain if needed, gives you the TXT record, verifies it and tells you where to forward email.
Receiving and sending are set up separately. From Pro, you set up sending for each sender address; you confirm ownership of your own domain via DNS. Changes in Command Mail do not automatically sync back to the original mailbox.
All agents use the same operations through hosted MCP with OAuth, local MCP, CLI or REST. Every session starts with capabilities. This shows the agent the owner, scopes, mailboxes with their autonomy levels and its limits.
Web and desktop: connect Command Mail from the Claude directory, where it is listed as a community connector. Or add a custom connector with the URL https://commandmail.app/mcp. Confirm access on the authorization page.
Command Mail is an official plugin in the ChatGPT plugin directory: open the plugin page, choose Install plugin and sign in to Command Mail. OpenAI says plugin availability depends on plan, workspace, role and region. Without plugin access, you can connect a custom MCP server with https://commandmail.app/mcp on the web if your plan, role and workspace allow it.
claude mcp add --transport http \
commandmail https://commandmail.app/mcpLocal MCP through the CLI: codex mcp add commandmail -- commandmail mcp. Or use the hosted URL if your version of Codex supports remote MCP servers.
EVE and your bots connect through the hosted MCP URL and work with the permissions you grant them.
Through the xAI API: add https://commandmail.app/mcp as a remote MCP server. Otherwise, use the CLI or REST.
Through the MCP URL once Muse supports custom connectors. Until then, use REST.
Use the hosted MCP URL if the runtime supports Streamable HTTP MCP. Otherwise, use the CLI or REST. This also applies to other agents.
More on permissions, workflow and limits: For agents.
The CLI is called commandmail. It works with scripts and scheduled tasks.
Install with one line, then log in. Requires Node.js 18 or later.
curl -fsSL https://commandmail.app/install.sh | sh
commandmail login| Command | Purpose |
|---|---|
commandmail login | log in |
commandmail mailboxes connect <adresse> | Connect an address |
commandmail domains claim <domain> | Claim your own domain |
commandmail domains verify <domain> | Verify a domain with a DNS TXT record |
Use the + next to the inbox tabs to create a custom view, such as “Customer conversations”. The view uses a shared label and includes archived emails with that label. Creating a view does not move any email.
Your agent receives our guide to organizing and archiving with the CLI installation. Through MCP, it reads the guide with get_organization_guide, or through the CLI with commandmail skill. It also explains how to handle your exceptions, false positives, invoices and narrowly defined filters.
commandmail skill
commandmail labels create --name "Kundenkommunikation"
commandmail labels list --json
commandmail inbox --label LABEL_ID --jsonA new view starts with manual assignments. Enable automatic labels in settings. Your agent suggests spam and archive rules based on concrete examples; you enable them under “Automation & rules”. Marking something as spam alone does not activate a rule.
In the morning brief, reported payment issues appear under “Review payments”, linked to the original email. A payment confirmation is not treated as an outstanding balance. Check the current balance with the provider; an email or risk score alone does not prove it.
Scopes define what a key or connected session can do. Grant only what the agent needs.
| Scope | Allowed |
|---|---|
mail:read | Read, search and view without marking as read; overview and morning brief; sort within the autonomy level; view activity and undo its own actions; suggest rules; follow up |
mail:draft | Save drafts. Never sends or overwrites your edits |
mail:send-approved | Send the exact version you approved |
mail:send | Send and reply independently from mailboxes on auto, each email with its own sendKey, within the daily limit (50 by default) |
mailboxes | Connect addresses, verify domain ownership with a TXT record, import old emails. Sending is set up for each sender address, from Pro |
Only a person in the browser can approve drafts, decide on suggestions, enable rules and change autonomy. Email content never grants permissions. Operator permissions such as admin are not available for customer keys.
Set the autonomy level for each mailbox: read → suggest → rules → auto.
| Level | What the agent can do |
|---|---|
read | Read. It changes no mail, not even read markers, and sends nothing, not even approved drafts. It can still save drafts; only you send them |
suggest | Archiving, spam and deletion actions are saved as suggestions (response {"suggested": true}) and wait for you. Sending works only through a draft you approve. Deleting for good and emptying Trash or Spam are not allowed |
rules | As with suggest: agent actions remain suggestions. Only rules you have enabled take effect. Here too it sends only approved drafts and deletes nothing for good |
auto | Actions run and are logged, with at most 50 conversations per call. With mail:send it sends on its own within the daily limit; with mail:read and mail:send it may delete for good. Supported sorting actions can be undone; sent and permanently deleted emails cannot |
Actions taken by agents and rules appear in the activity log and can be undone. Suggestions wait there for your decision.
For mailboxes such as customer support, a suitably configured agent can answer questions and send replies independently. It knows your offers, agreements and tone only from the information and sources you make available to it. Regular execution is configured separately in the agent.
mail:send. Agents that only sort email do not need it.auto. In settings under Automation. Below it, the agent saves its replies as drafts for you to approve.Each email uses its own sendKey. Repeating the same request with the same key refers back to the same send attempt. An agent does not answer detected automated messages (RFC 3834, Precedence: bulk, mailing lists, out-of-office replies, bounces) on its own. It does not write on its own to no-reply addresses or to senders that only ever sent automated mail; it saves a draft instead, which you can approve.
The morning brief summarizes what arrived since yesterday and what is due by tomorrow:
Expand it to the week with one button. Agents retrieve it with morning_brief.
Each email receives percentage estimates for spam, fraud and phishing. The assessment reads text only. It never opens links or attachments.
Alongside risk assessment, Command Mail sorts email into General, Social, Promotions, Updates and Forums, and adds markers such as Appointment, Invoice, Deadline and Security.
No. Signing in with Google shares only your name, email address, profile picture link and account ID. Command Mail sees the emails you forward to your private inbound address.
Yes, if you grant it mail:send and the mailbox is set to auto. Every email is logged and the daily limit applies (50 by default). Without this permission, it saves drafts and sending follows your approval. How to set it up
Reloading does not create a new send attempt for the same approved draft version. If the outcome is unclear, check the recorded send status before starting a new send.
Never on its own: automated emails (RFC 3834, Precedence: bulk, mailing lists, out-of-office replies) and no-reply addresses get no agent mail without your approval.
An installable web app with web push, plus Telegram notifications. Desktop apps for Mac and Windows are planned.
German, English, Spanish, French and simplified Chinese.
You can start free with one address. Pro costs €19 per month as a founder price for the first 100 paying Pro accounts, as long as the subscription runs, and €29 per month for all later accounts; yearly billing gives you two months free. You can try Pro for 7 days for €1. Team costs €39 per person per month; Enterprise is priced on request. All prices include VAT. See pricing
Command Mail is a product of Fyn Labs LLC. Mathias Heinke answers inquiries and handles personal setup.