Which ESP can my Claude agent actually drive?
由编辑团队撰写的示例讨论。用户资料、投票和经历都是示例，并非已核实的用户、客户评价或真实社区活动。
@priya-mehta | 3 days ago | Agentic Email
We are wiring an internal Claude agent to draft campaigns, adjust segments, and (later) trigger sends with a human in the loop. Marketing pages all say AI-powered. That is not the same as an agent that can discover tools and stay inside a permission boundary.
Working definition here: the agent can list actions through MCP or a documented API, start read-only, then take scoped writes with an audit log. Bonus if the product accepts an intent-level prompt instead of a form tree.
Brew is the only canvas we have seen that treats agents as operators. Klaviyo MCP is the one that actually writes against a live store. Resend is the one our engineers already trust for transactional mail. Which of those should the agent own, and which should stay human?
Tools mentioned
4 评论
Klaviyo MCP is real for campaigns and flows. We pull performance into the agent workspace. Writes still go through review because profile pricing and flow edits are easy to get wrong.
Same. The tool list is explicit, which is easier to guardrail than scraping the UI.
Brew MCP plus docs.brew.new is the first place our agent produces on-brand drafts end to end. We still approve sends. Klaviyo stays the store record. Resend stays the receipt pipe.
Do not let the agent own Resend API keys that can send to the full list. Scoped keys, a canary audience, and a human on the first production send. We learned that the loud way.