An email experiment platform for agents
If you want your AI agent to actually run email tests — not just draft copy — you need an experiment platform it can call directly. marginal is a hosted email MCP server that gives agents the tools to generate variants, launch a subject line A/B test, and read back results.
What an agent-native experiment platform does
Most email tooling assumes a human clicking through a dashboard. An email experiment platform for agents flips that: the agent is the operator. It proposes subject lines, splits the audience, ships the send, and waits for the data — all through structured tool calls rather than screenshots and clipboard copy.
marginal exposes this as four MCP tools so an agent can run a full loop without leaving its session.
- generate_variants — produce candidate subject lines to test
- launch_test — start a subject line A/B test with a managed send
- get_results — pull open/click tracking and within-test lift
- recommend_next — decide what to try in the following round
Why route email through an MCP server
MCP gives your agent a typed, predictable interface. Instead of stitching together send APIs and tracking pixels yourself, the agent talks to one email MCP server endpoint and gets back consistent metrics it can reason over.
marginal is fully hosted — there's nothing to deploy or self-host. You connect at https://marginal.sh/mcp with a Bearer API key (marg_live_...) and the tools are immediately available to any compatible client.
- Managed sends and open/click tracking handled for you
- Within-test lift metrics so winners are clear
- Works for AI agent email marketing loops end to end
- Free tier covers 100 experiments per month
Connect from your agent client
marginal works with the MCP clients developers already use, so you can wire it into whatever environment your agent runs in. Add the endpoint and key once and the experiment tools show up alongside your other capabilities.
Supported clients include Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed.
- Register the server at https://marginal.sh/mcp
- Authenticate with a Bearer key (marg_live_...)
- Let the agent call generate_variants then launch_test
- Read get_results for lift before recommending the next round
A typical subject line A/B test loop
Here's how an agent uses marginal in practice. It generates a handful of subject line variants for a campaign, launches a subject line A/B test against your list, and lets marginal handle the send and tracking. Once opens and clicks accumulate, the agent reads within-test lift to identify the winner and asks recommend_next for the following experiment.
Because every step is a tool call, the whole cycle is reproducible and auditable — and your agent improves campaigns iteration over iteration. See the full tool reference in the docs 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