Email Marketing for AI Agents
AI agents are great at drafting copy, but they go quiet the moment you ask them to actually send an email and learn what worked. marginal closes that gap with a hosted email MCP server that lets your agent run subject line experiments, send campaigns, and read back the numbers — all over the Model Context Protocol.
Why agents need a dedicated email layer
Most agent setups stop at generating text. Email marketing for AI agents means going one step further: launching the send, tracking opens and clicks, and feeding the outcome back into the next decision. That loop is hard to wire up by hand and easy to get wrong.
marginal exposes that loop as a small set of MCP tools. Your agent calls them the same way it calls any other tool, so the model can propose copy, test it, and act on the result without you stitching together separate sending and analytics services.
- generate_variants — produce subject line and body options to test
- launch_test — start a managed send with a subject line A/B test
- get_results — pull open/click tracking and within-test lift
- recommend_next — turn results into a concrete next action
How subject line A/B testing works
A subject line A/B test is the highest-leverage experiment in email, and it's exactly the kind of repetitive optimization agents handle well. With marginal, the agent generates candidate subject lines, launches a split test against your list, and waits for engagement data to come back.
Once results arrive, within-test lift metrics tell the agent which variant won and by how much — not just a raw open rate, but the relative gain over the control. That signal is what recommend_next uses to suggest the follow-up send or the next round of variants.
- Split sends automatically across variants
- Track opens and clicks per variant
- Compare lift inside the same test for a fair read
- Let the agent iterate without manual spreadsheet work
A hosted MCP server, not a side project
marginal is a hosted email MCP server — there's nothing to deploy or self-host. You connect at https://marginal.sh/mcp using a Bearer API key (marg_live_...), and the sending infrastructure, tracking, and experiment storage are managed for you.
It works with the MCP clients your team already uses, so AI agent email marketing fits into your existing workflow rather than forcing a new one. The free tier covers 100 experiments per month, which is plenty to validate the approach before scaling up.
- Endpoint: https://marginal.sh/mcp · Registry: sh.marginal/mcp
- Works in Cursor, Claude Desktop, Claude Code, Codex, Windsurf, Cline, Continue, and Zed
- Free tier: 100 experiments/month
- Docs and setup details at https://marginal.sh/docs/
A typical agent workflow
Put together, the tools form a clean optimization cycle. The agent drafts options, tests them on real recipients, reads the lift, and decides what to do next — repeating until the metric stops improving.
Because every step is an MCP tool call, you can describe the goal in plain language and let the agent drive the experiment end to end.
- Ask the agent to improve open rates for a campaign
- It calls generate_variants for fresh subject lines
- launch_test runs the managed split send
- get_results returns engagement and lift; recommend_next plans the follow-up
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