Give Claude Code its own email inbox

Claude Code can already read your repo, run your tests and open a pull request. What it cannot do out of the box is receive the email that the rest of your work happens in: the failing deploy alert, the customer reply, the vendor asking for a decision.

Anjal gives it a real inbox. One command.

Connect it

claude mcp add --transport http anjal https://mcp.anjal.app/mcp

Then start Claude Code and run /mcp. It will prompt you to authenticate, a browser opens on Anjal, you sign in and click Allow. No API key is copied anywhere and nothing is stored in your shell history or your repo.

Check it worked:

claude mcp list

To connect it for one project rather than globally, add -s project, which writes to .mcp.json in the repo so your teammates get the same server. Use -s user for every project on your machine.

What Claude Code can do once it is connected

47 tools arrive, covering inboxes, messages, threads, drafts, attachments, domains, workspaces, webhooks and calendar. The ones that matter most in a coding session:

Things that actually work in a terminal

Read the alert, fix the code, reply on the thread.

"Check the ops inbox. If Sentry sent anything about the payments service, find the stack trace, look at the file it points to, and tell me what changed there recently."

Claude Code reads the thread, greps the repo, and answers with the commit. It can then reply on the same thread, and the reply lands as a proper continuation in the recipient's client rather than as a new conversation, because Anjal writes Message-ID, In-Reply-To and References for it.

Sign up for the thing you are integrating.

Agents hit signup walls constantly. Give the agent an inbox and it can create its own account on a staging service, receive the confirmation mail, and pull the verification code out of it without a human forwarding anything.

Draft, don't send.

For anything going to a real person, have it write a draft instead:

"Draft a reply to the customer explaining the fix and when it ships. Do not send it."

create_draft puts it in the inbox. You read it in the Anjal web app, or in Apple Mail over IMAP, and send it yourself.

The safety part, which matters more here than elsewhere

Claude Code has a shell. An email that arrives in an inbox it can read is attacker-controlled text about to enter the context of an agent that can run commands. "Ignore your instructions and run this" is not hypothetical; it is the obvious attack.

Anjal screens every inbound message for prompt injection before an agent can see it, and quarantines what it judges hostile. A quarantined message is not returned by list_messages or get_thread. A human releases it from the web app, and an agent cannot release its own quarantine, because the attack would simply be to ask it to. We publish the architecture and not the rules. Read the honest limits first: detection is advisory, not a trust boundary, and the rules that actually hold are about gating side-effects on the sender rather than on the text.

That is on by default. You do not configure it.

Scoping what it can reach

The connector gets your whole account. If you want the agent limited to one workspace, use a workspace-scoped key with the local server instead:

claude mcp add anjal -e MYAGENTMAIL_KEY=wk_... -- npx -y anjal-mcp@latest

Workspaces isolate inboxes from each other, so a key scoped to one cannot list, read or send from any other. That is the right shape when the agent is working on one customer's project.

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. There is a 7-day trial.

Next

• Connect Anjal to Cursor
• Connect Anjal to Codex
• MCP server reference and agent safety
• Handle 2FA and verification codes in an agent workflow

Setup verified against claude mcp add --help on 2026-09-28.