The Model Context Protocol Email Server for AI Agents
marginal is a hosted Model Context Protocol email server that lets AI agents draft subject lines, launch experiments, and read results — without leaving your editor or agent loop.
What a Model Context Protocol email server does
The Model Context Protocol (MCP) gives AI agents a structured way to call external tools. A Model Context Protocol email server exposes email marketing actions — generating copy, sending campaigns, and reading metrics — as MCP tools your agent can invoke directly.
marginal is that server, fully hosted. You point any MCP-compatible client at https://marginal.sh/mcp, authenticate with a Bearer API key, and your agent gains the ability to run real subject line A/B tests against a live audience.
- generate_variants — produce subject line candidates from a prompt
- launch_test — start a managed A/B send
- get_results — pull open, click, and lift data
- recommend_next — get a data-backed suggestion for the next experiment
Why teams choose an email MCP server over a raw API
Wiring an agent to a traditional email API means writing glue code, handling auth refresh, and parsing responses by hand. An email MCP server removes that work: the tools are already described in a format your model understands, so the agent reasons about them natively.
Because marginal is hosted, there is no self-hosted infrastructure to maintain. Managed sends, open and click tracking, and within-test lift metrics all happen on our side, and your agent simply reads the outcome.
- No backend code to deploy or babysit
- Consistent tool schemas across every supported client
- Within-test lift metrics so winners are obvious
- Free tier covers 100 experiments per month
Running a subject line A/B test through your agent
A typical AI agent email marketing flow looks like a short conversation. You ask the agent for ideas, it calls generate_variants, then launch_test starts the managed send. Once enough opens and clicks accumulate, get_results returns the lift and recommend_next proposes a follow-up.
This loop keeps a human in the driver's seat for strategy while the Model Context Protocol email server handles the mechanics of the subject line A/B test.
- Ask for variants, then approve which ones ship
- Launch a managed test without leaving your editor
- Track opens and clicks in near real time
- Iterate using recommend_next instead of guessing
Supported clients and getting started
marginal works with the major MCP clients, so the same Model Context Protocol email server backs whichever tool your team prefers. Add the endpoint and your marg_live_ key, and the tools appear automatically.
Supported clients include Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed. Full setup details live in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...)
- Registry: sh.marginal/mcp
- Docs: 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