An email variant testing platform built for AI agents
marginal is an email variant testing platform delivered as a hosted MCP server. Instead of clicking through a dashboard, your AI agent generates variants, launches tests, and reads lift metrics by calling four MCP tools.
What an email variant testing platform actually does
A variant testing platform exists to answer one question reliably: which version of an email performs better? marginal handles the full loop — drafting candidate variants, sending them to split audiences, tracking opens and clicks, and computing within-test lift so you act on signal instead of hunches.
What makes marginal different is the surface area. The same workflow that normally lives behind a web UI is exposed through an email MCP server, so an AI agent can drive experiments end to end without a human translating intent into clicks.
- generate_variants — produce subject line or body candidates to test
- launch_test — split your audience and send the variants
- get_results — pull open/click rates and within-test lift
- recommend_next — get a data-backed suggestion for the next iteration
Running a subject line A/B test
Subject lines are the highest-leverage thing to test because they gate every open downstream. With marginal, a subject line A/B test starts by generating a handful of variants, launching them against a segment, and waiting for tracked opens to accumulate.
Because lift is measured within the test, you compare variants under the same send conditions — same audience, same window — rather than against a stale historical baseline. That keeps your conclusions honest and repeatable.
- Test 2–5 subject lines per experiment for clean attribution
- Open tracking surfaces which line earned attention
- Click tracking confirms the open translated to intent
- Within-test lift quantifies the winner's margin
Why AI agent email marketing fits this model
AI agent email marketing works best when the agent can both reason about copy and execute the experiment. marginal connects to MCP clients like Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Codex, and Zed, so the agent you already use can run tests in your workflow.
Authentication is a single Bearer API key (marg_live_...) against the hosted endpoint at https://marginal.sh/mcp. There is no self-hosted server to maintain — marginal runs the sends, tracking, and metrics for you. The free tier covers 100 experiments per month, which is plenty to validate the loop before scaling.
- Hosted endpoint: https://marginal.sh/mcp
- Registry name: sh.marginal/mcp
- Docs and setup: https://marginal.sh/docs/
- Free tier: 100 experiments/month
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