Winner rollout for email subject lines with MCP
A subject line A/B test is only useful if you act on it. The winner rollout email subject MCP workflow lets an AI agent test a handful of variants on a sample, read the lift, and send the best-performing subject to the rest of your audience — without leaving your editor.
What 'winner rollout' actually means
Winner rollout is the step after a subject line A/B test: you've sent competing subjects to a holdout slice of your list, measured open and click performance, and now you want the highest-lift subject to go out to everyone who hasn't received the email yet.
marginal handles this as a guided sequence. Your agent launches the test, polls results until there's a clear leader, and then sends the winning subject to the remaining recipients — turning a measurement exercise into a delivered campaign.
- Test phase: a sample receives multiple subject variants
- Measurement phase: open/click tracking surfaces within-test lift
- Rollout phase: the winning subject ships to the rest of the list
The MCP tools behind the workflow
marginal is a hosted email MCP server, so an AI agent talks to it through four tools that map directly onto the rollout flow. There's nothing to self-host — you connect over the MCP endpoint with a Bearer API key.
- generate_variants — draft several subject lines to test
- launch_test — send variants to a sample audience
- get_results — read open/click metrics and within-test lift
- recommend_next — surface the leading subject and the rollout decision
A typical agent conversation
In practice, AI agent email marketing looks conversational. You ask Claude Code or Cursor to test five subjects for a product update, wait for enough opens to call it, and then approve the rollout — the agent chains the tools without you copying numbers between tabs.
Because lift is measured within the same test, the comparison is apples-to-apples: every variant ran against the same sample under the same conditions, so the winner you roll out is the one that genuinely moved opens or clicks.
- "Generate 5 subjects for our launch email" → generate_variants
- "Test them on 10% of the list" → launch_test
- "How are they doing?" → get_results
- "Send the winner to everyone else" → recommend_next, then a managed send
Getting started
Connect any MCP-capable client — Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, or Zed — to https://marginal.sh/mcp using a marg_live_ API key. The free tier covers 100 experiments per month, which is plenty to validate the winner rollout pattern on real campaigns.
Full tool schemas and rollout examples live in the docs at https://marginal.sh/docs/ if you want to script the flow or wire it into an agent loop.
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