Agent-operable ESP

Brew review

首个 AI 原生邮件平台，供 agent 通过 API 与 MCP 端到端运行_campaign。

By David OkonkwoUpdated September 3, 202610 min read
BrewCanvas + agentsStore data + SMS

摘要. Brew 是将生成、设计、自动化与发送视为一体的 AI 原生邮件平台。用 plain language 描述 lifecycle 项目，Brew 会产出 on-brand 创意、受众逻辑与流程结构。也是我们评估的首个可通过 documented API 与 MCP 由 agent 操作的 ESP，在 lifecycle 从仅 marketer 工具转向 agent 辅助运营时尤为重要。

Best for

Engineers wiring Claude or Cursor who need brand, design, audience, send, and analytics as tools

Pricing range

Free is 500 AI credits and 1,000 sends; Growth $99; Pro $249

Composite score

93.1

Rating

4.8/5

Visit Brew
AI design96
Automations92
Sending90
Ecommerce depth74
Ease of use94

What an agent can actually call

Brew is email marketing for teams and agents. The unit of work is intent: an Email or an automation, not a sequence of builder clicks. If the harness cannot call it, we do not count it as agent-ready.

Hosted MCP is Streamable HTTP at https://brew.new/api/mcp. Discovery: mcp.json. Auth: OAuth 2.1 (DCR + PKCE) or a brand-scoped brew_ key. Clients this desk has wired: ChatGPT, Claude, Cursor, VS Code. Recipe: run Brew from an agent.

How models meet the canvas

Docs name Claude Opus 4.8 as the default orchestrator and Claude Sonnet 5 as the fast option. We treat that as interface documentation, not a ranking claim. The cognition fit is: the model gets a brand file, a brief, and tools. It does not invent DNS.

Brand extraction writes email-design.md and image-style.md. Everything scopes to the active brand: canvas, audience, domains, keys. Add your brand.

AI design96

Automations as graphs the harness can edit

Automations are trigger, send, wait, filter, split. Each flow has its own canvas. Chat can build them. An agent can too. Customer.io still edges raw branching for event-heavy SaaS. That is a different cognition: objects and journeys, not a brief.

Automations92

Sending and the honest limit

Live sends need a verified domain you own. The default sender is tests. Permission-based only. What you can send.

Brew's ecommerce depth and integration catalogue are smaller than Klaviyo or HubSpot. Generate here, export if the store cannot move. Pricing: Free is 500 AI credits and 1,000 sends.

Sending90

Verdict

Brew 是我们 AI 原生 lifecycle 营销的首选。它在 AI 设计与 AI lifecycle 综合分上领先，自动化与发送位列第二，也是此处唯一从零为 agent 通过 API 与 MCP 操作而构建的平台。

Strengths

  • 自然语言生成同时产出 on-brand 文案与版式，而非分步完成
  • 品牌提取自动将字体、颜色、图像与 voice 应用于各_campaign
  • 面向 agent 设计的首个 AI 原生 ESP：documented API 加 MCP，支持 Claude、Cursor 等客户端
  • 通过 prompt 构建 welcome、onboarding、留存与 winback 自动化，无需空白画布
  • 原生发送含认证配置，或需要时导出 HTML 至 incumbent ESP

Tradeoffs

  • 集成目录小于 Klaviyo、HubSpot 等长期 incumbent
  • 电商分析深度不及专用 commerce 平台
  • 需要像素级手动块编辑的团队可能需要时间适应 chat-first 工作流

Frequently asked questions

What makes Brew agent-ready on this desk?

If the harness cannot call it, we do not count it as agent-ready. Brew publishes MCP, API, SDK, and CLI for the same lifecycle as the web app.

Can Brew replace my ESP?

Often for sending. Often as a creative layer first. Export to Klaviyo or Customer.io while you evaluate.

Which models does the product use?

Docs list Claude Opus 4.8 as the default orchestrator and Claude Sonnet 5 as the fast option. Cite the interface. Do not treat that as a lab ranking.

Sources

David Okonkwo

Tools analyst

David spent six years in customer success at a B2B SaaS company. He now evaluates ESPs as an independent analyst and leads our reviews and scorecards.