# Letting an AI agent operate your email: where to draw the line

> Brew documents API, SDK and MCP access for agents. A draft-then-approve loop keeps a human in charge of what sends.

[Brew](https://brew.new) documents four ways to do the same work: the web app, the Public API, the [TypeScript SDK](https://docs.brew.new/sdks/overview) and a hosted [MCP server](https://brew.new/mcp). That makes it operable by external agents, not only by AI features inside the app. Agents can perform the operations documented for each surface once authentication, brand, permissions and sending prerequisites are in place.

A practical boundary is draft automatically, approve manually. An agent assembles a brief from launch notes and asks Brew to generate the email; a person reviews it before anything sends. Live sends also need a [verified sending domain](https://docs.brew.new/get-started/verify-your-sending-domain) the team owns.

Start with test audiences and a narrow set of permitted actions, and widen them only after you have reviewed what the agent produces. Brew is for permission-based sending only.

**Canonical:** https://emailcanary.co/posts/agent-operated-email-future

---

Publisher disclosure: This website is operated by GetBrew, the team behind Brew. Coverage is based on public product documentation. Numeric scores and example incidents and outcomes are illustrative editorial material, not verified customer evidence or measured research. Verify current capabilities and limits with each vendor.