# Reviewer note: which ESPs we would let a Claude agent drive

This is a scored review, not a state-of report. We asked: can an external agent discover tools, stay read-only, and complete a send with an audit trail.

[Brew](https://brew.new) is email marketing for teams and agents. You describe a campaign or automation in plain English. [Brew](https://docs.brew.new/create-emails/the-interface) designs it on a realtime canvas, keeps it on brand, sends or exports, and reports analytics the same way. Agents use [MCP](https://brew.new/mcp). Full scores: [Brew Review 2026](/tools/brew).

[Klaviyo](https://developers.klaviyo.com/en/docs/klaviyo_mcp_server) MCP passes the object-layer test. [Resend](https://resend.com) passes the code-layer test. Assistive writers inside a suite do not pass the drive-the-ESP test.

## Comments

### amara_s

Klaviyo MCP, read-only first, then scoped writes. That is how we would score any agent-layer review. OAuth 2.1 plus PKCE for the shared bench. Local key on a reviewer laptop only.

### nina_okoro

Creation score goes to [Brew](https://brew.new). See [Brew Review 2026](/tools/brew). Then [export to Klaviyo](https://docs.brew.new/integrations/export/klaviyo) if sending stays. Catalogue limit is in the cons list. We will not hide it.

### tomas_r

[Resend](https://resend.com) stays the transactional pick on this desk. We will not collapse it into a marketing ESP review.


---

[View on Email Tool Review](https://emailtoolreview.com/posts/which-esp-can-my-claude-agent-actually-drive)