Give Devin an email inbox

Devin works for hours without you watching. That is the whole pitch, and it is also why email matters more here than for an editor assistant: when a run ends at 3am, or stalls on a service that wants an account, there is nobody at the keyboard.

An inbox gives Devin a way to tell you, and a way to get itself unstuck.

Connect it

Two paths, and which one you use depends on whether you are an org admin.

From the UI, which an organization admin does once for everyone: go to Customize → MCPs, click Add custom MCP, and add https://mcp.anjal.app/mcp as a remote HTTP server. Devin supports OAuth for remote servers, so it registers itself and you sign in at Anjal and click Allow.

From the CLI:

devin mcp add anjal https://mcp.anjal.app/mcp

If you do not see the Add custom MCP button, you are not an org admin. Devin gates this deliberately. Use Suggest MCP Integration to request it rather than trying to work around it, and send your admin this page.

This is the main way Devin differs from Claude Code or Cursor: connecting a tool is an organisation decision, not an individual one. That is a feature for a product that runs unattended, and it means the useful conversation is with whoever administers your Devin org.

What Devin gets

47 tools. For unattended work, these earn their place:

Three things this unlocks

A run that ends can tell you, and you can answer. Devin sends the summary from its own address. You reply from your phone. On the next run it reads your reply with get_thread and continues. That is a far better loop than a log you have to go and find, and unlike a Slack notification it has a thread.

It gets past signup walls on its own. Integrating a third-party API usually means creating an account on it and clicking a confirmation link. With an inbox, Devin registers, receives the mail and reads the verification code without a human forwarding anything from their personal inbox.

Nothing reaches a customer without a person. For anything outbound to a real human, have it call create_draft rather than send_message. The draft sits in the inbox until you read it in the Anjal web app, or in Apple Mail over IMAP, and send it. For an agent working unattended for hours this is the difference between a useful colleague and an incident report.

Scope it down, because it is unattended

The OAuth connector reaches your whole Anjal account. For an autonomous agent that is more than it needs. Give Devin a workspace-scoped key and the local server instead:

devin mcp add anjal --stdio "npx -y anjal-mcp@latest" --env MYAGENTMAIL_KEY=wk_...

A wk_ key reaches one workspace and cannot list, read or send from any other, or see that any others exist. Of everything on this page, this is the recommendation I would push hardest: an agent that runs for hours without supervision should hold the narrowest credential that still does the job.

The risk that is specific to autonomy

Inbound mail is text written by strangers entering the context of an agent that executes code unsupervised. "Ignore your previous instructions" in a support email is the obvious attack and it is not theoretical.

Anjal screens every inbound message on the ingest path and quarantines what it judges hostile, so the agent never reads it, and an agent cannot release its own quarantine, because the attack would be to ask it to. We publish the architecture and not the rules. Read the honest limits before relying on it: detection is advisory, not a trust boundary, and gating side-effects on the sender in code is what actually holds.

Cost

Starter is $5/month for 2 inboxes and 6,000 emails. Developer is $15/month for 10 inboxes and 20,000 emails. Extra inboxes are $2/month each. 7-day trial.

Next

• Connect Anjal to Codex, the other long-running one
• Connect Anjal to Claude Code
• MCP server reference and agent safety

Setup verified against Devin's MCP documentation on 2026-09-28. Devin's CLI flags change; if devin mcp add differs for you, the UI path is the stable one.