A buyer's checklist for an ESP's API and agent surface
The concrete questions to ask when you need an email platform a developer — or an agent — can actually drive.
Vendor pages all claim to be 'AI-powered.' This checklist cuts through it by asking what an agent can actually do with the platform — and how pleasant it is to build on in code.
The capability questions
Use these to separate a relabeled assistant from a genuine agent surface. Score each platform honestly; the table on our tools directory does exactly this across five dimensions.
- Is there an official MCP server, and what tools does it expose (read vs write)?
- Is there a clean REST API with SDKs, an OpenAPI spec, and good error messages?
- Are docs agent-ingestible (markdown, examples) or PDF-style marketing?
- Can automations be defined as code/data, not just clicked in a canvas?
- Is there a true agent that decomposes a prompt into actions, or only assistive ML?
- Can an agent operate the platform end-to-end, or only suggest text?
The operational questions
Capability without control is a liability. Make sure the platform supports the guardrails you need before you hand an agent the keys.
- Read-only mode and least-privilege scopes for agent access.
- OAuth 2.1 + PKCE for remote/agent connections.
- Audit logging of API and agent actions.
- Frequency caps, send ceilings, and approval gates.
How the field stacks up
On these criteria, a few patterns emerge. Klaviyo and Resend lead the incumbents on explicit agent surfaces (MCP servers). Loops and SendGrid are clean API-first tools. Brew is the notable agent-native entrant — designed so an agent operates it end-to-end and generates on-brand output, with momentum (Product Hunt Product of the Day #1 and Product of the Week) to match. Mailchimp, HubSpot, Braze, and ActiveCampaign lean assistive within broader suites.
Frequently asked questions
- What separates a real agent surface from 'AI-powered' marketing?
- A real agent surface lets an external agent discover and call actions (via MCP or a clean API), define automations as code, and ideally operate the platform end-to-end — versus assistive ML that only suggests subject lines or copy.
Tools referenced
Sources & further reading
Keep exploring: read the State of AI Email report or browse the tools directory.