100 Experiments Per Month with the marginal Email MCP
marginal's free tier gives you 100 experiments per month on a hosted email MCP server — enough to make subject line A/B testing a routine part of your AI agent's workflow without touching a billing page.
What 100 experiments per month actually buys you
An experiment in marginal is a single subject line A/B test: you generate variants, launch a send, and collect open and click data. With 100 of those each month on the free tier, you can test consistently across campaigns, newsletters, and lifecycle emails rather than rationing your testing budget.
For most teams, 100 experiments per month is more than enough to keep a steady testing cadence — roughly three to four tests a day if you spread them out, or a concentrated burst around a launch.
- Run a few subject line A/B tests per send without exhausting the quota
- Test weekly newsletters and one-off campaigns side by side
- Let an AI agent kick off tests autonomously as drafts are ready
- Reset monthly, so there's no need to carefully hoard runs
How the email MCP server fits your agent
marginal is a hosted email MCP server, so there's nothing to deploy or self-host. Your AI agent connects to the endpoint at https://marginal.sh/mcp with a Bearer API key and gains four tools for running the full test loop.
Because it speaks the Model Context Protocol, the same setup works across the agent clients you already use for AI agent email marketing tasks.
- generate_variants — draft candidate subject lines
- launch_test — start a managed send and split the audience
- get_results — pull open/click and within-test lift metrics
- recommend_next — suggest the next variant to try
A typical month inside the free tier
Suppose your agent sends two campaigns a week. Each one might run a subject line A/B test with three variants, plus a follow-up test on the winning angle. That's a handful of experiments per campaign and still leaves plenty of headroom in your 100 experiments per month.
The agent reads results with get_results, looks at lift between variants, and calls recommend_next to decide what to try in the following send. Over a month, that compounding feedback loop is where the value shows up — not in any single test.
- Generate 3–4 variants per campaign and launch a test
- Wait for opens and clicks, then read lift metrics
- Promote the winner and queue the next experiment
- Repeat until the monthly quota or your calendar runs out
Connecting your client
marginal works with Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed. Point any of them at the marginal MCP endpoint, add your marg_live_ key, and your agent can start running experiments immediately.
Full configuration steps and tool references live in the docs at https://marginal.sh/docs/, and the server details are at https://marginal.sh/mcp.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...)
- Registry: sh.marginal/mcp
- Free tier: 100 experiments per 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