A self serve email experiment platform for AI agents
marginal lets you spin up subject line tests, send to your list, and read lift metrics without waiting on a marketing ops queue. It's a self serve email experiment platform that AI agents drive directly through the Model Context Protocol.
What self serve means here
Most email tools assume a human clicking through a dashboard. marginal flips that: your agent calls a hosted email MCP server and runs the full experiment loop on its own. There's nothing to install and nothing to host — point a client at the endpoint, authenticate, and start testing.
Self serve doesn't mean unguided. The toolset is small and opinionated so an agent can reason about what to do next without you stitching together five APIs.
- Hosted at https://marginal.sh/mcp — no self-hosted server to maintain
- Bearer API key auth using a marg_live_ token
- Free tier covers 100 experiments per month
- Works as soon as your MCP client connects
The experiment loop, end to end
A complete subject line A/B test is four tool calls. An agent generates candidate subject lines, launches the test against your audience, waits for engagement, and asks marginal what to do with the result.
- generate_variants — draft subject line options for a campaign
- launch_test — send the variants as a managed A/B test
- get_results — pull open and click tracking with within-test lift
- recommend_next — get a data-backed call on the winner or next step
Why agents fit this workflow
AI agent email marketing works best when the agent owns the iteration, not just the copy. With marginal, the same agent that drafts variants can launch the subject line A/B test, read the lift, and propose the next round — closing the loop without a handoff.
Because the tools return structured results, the agent can compare variants quantitatively instead of guessing. That makes the experiment platform genuinely self serve: you describe a goal, the agent runs the test and reports back.
Connect your client
marginal speaks MCP, so it plugs into the editors and agent runtimes you already use — Cursor, Claude Desktop, Claude Code, OpenAI Codex, Windsurf, Cline, Continue, and Zed. Add the endpoint and your key, and the four tools appear to the agent automatically.
Full setup and tool reference live in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Registry: sh.marginal/mcp
- Compatible with every major MCP client
- Same tools regardless of which client you launch from
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