The Developer Email Marketing Stack
Most email tooling is built for marketers clicking around a dashboard. If you live in an editor and orchestrate work through AI agents, you need a developer email marketing stack that speaks your language: APIs, structured results, and tools your agent can actually call. marginal fills that gap with a hosted email MCP server.
What belongs in a developer-first stack
A traditional marketing stack assumes a human composing campaigns. A developer stack assumes code and agents driving the same outcomes — generating variants, launching experiments, and reading lift metrics without leaving the terminal.
marginal slots in as the experimentation and send layer. It exposes four MCP tools so your AI agent can run a full subject line A/B test loop programmatically.
- generate_variants — produce candidate subject lines to test
- launch_test — start a managed send across variants
- get_results — pull open/click tracking and within-test lift
- recommend_next — get a data-backed next step
Why an email MCP server changes the workflow
MCP (Model Context Protocol) lets your editor's agent discover and call tools directly. Instead of copy-pasting subject lines into a separate app, the agent calls the email MCP server inline as part of whatever task you're working on.
Because marginal is hosted at https://marginal.sh/mcp, there is nothing to deploy or maintain. You authenticate with a Bearer API key (marg_live_...) and your agent has experimentation capabilities immediately. There is no self-hosted server to run.
- Connects to Cursor, Claude Code, Claude Desktop, Codex, Windsurf, Cline, Continue, and Zed
- Registry identifier: sh.marginal/mcp
- Managed sends with built-in open and click tracking
Running subject line A/B tests with an AI agent
The core loop is simple and well-suited to AI agent email marketing. Ask your agent to draft variants, launch the test, and report back the winner once enough data lands.
Within-test lift metrics tell you which subject line is pulling ahead and by how much — so the agent's recommend_next call has real numbers behind it rather than a guess. You stay in your IDE the whole time.
- Generate several subject line candidates from a prompt
- Launch a managed A/B test against your audience
- Read open/click results and within-test lift
- Act on recommend_next to iterate the next round
Getting started for free
The free tier includes 100 experiments per month — enough to validate the workflow and run real subject line A/B tests across a project. Point your MCP client at the endpoint, add your API key, and start calling tools.
Full setup steps and tool references are in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Free tier: 100 experiments/month
- 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