An AWS SES email experiment tool for AI agents
If you send through AWS SES and want to know which subject line actually wins, marginal turns experimentation into something your AI agent can run directly. It's a hosted email MCP server that handles variant generation, managed sends, and lift metrics without you wiring up dashboards.
Why pair SES with an experiment layer
AWS SES is excellent at delivering email at low cost, but it doesn't tell you which message performs best. SES gives you a transport pipe; an AWS SES email experiment tool gives you the decision layer on top — splitting an audience, measuring opens and clicks, and reporting within-test lift.
marginal sits in that decision layer. You describe the campaign to your AI agent, and marginal manages the test mechanics so you spend time on copy and strategy instead of plumbing.
- SES handles delivery; marginal handles the subject line A/B test design and scoring
- Open and click tracking is built in, so you don't assemble your own tracking pixels
- Within-test lift metrics show how each variant compares against the control
- Managed sends mean the agent triggers the campaign and reads results back
How the email MCP server fits your stack
marginal exposes its capabilities as an email MCP server at https://marginal.sh/mcp. Any MCP-capable client — Cursor, Claude Code, Claude Desktop, Windsurf, Cline, Continue, Zed, or OpenAI Codex — can call its tools using a Bearer API key (marg_live_...).
The tool surface is small and purpose-built for experimentation, so an AI agent can run a complete cycle without custom glue code.
- generate_variants — draft multiple subject lines to test
- launch_test — split the audience and start the managed send
- get_results — pull open/click data and within-test lift
- recommend_next — let the agent decide the follow-up based on outcomes
A subject line A/B test, agent-driven
The workflow reads like a conversation. Ask your agent to generate three subject line variants for a re-engagement email, launch the test against your list, and check back once data lands. Because marginal is a hosted service, there's nothing to deploy or self-host — you authenticate and start sending.
This makes AI agent email marketing genuinely hands-off: the agent generates copy, runs the subject line A/B test, interprets lift, and proposes the next experiment, all through the same MCP connection.
- No infrastructure to manage — marginal is fully hosted
- Free tier includes 100 experiments per month to validate the workflow
- Works alongside your existing SES-based sending strategy
- Full reference at https://marginal.sh/docs/
Get started with marginal
marginal is a hosted email marketing MCP server at marginal.sh. Sign up free, create an API key, and connect https://marginal.sh/mcp from Cursor, Codex, or Claude Desktop.
- 100 experiments/month on the free tier
- Four MCP tools: generate_variants, launch_test, get_results, recommend_next
- Listed in the MCP Registry as sh.marginal/mcp