How to Compare Subject Lines in One Send with marginal
Testing two or more subject lines shouldn't require juggling spreadsheets or guessing which line won. This how-to guide walks through comparing subject lines in one send using marginal — the hosted email MCP server your AI agent can drive end to end.
What you need before you start
marginal exposes four MCP tools — generate_variants, launch_test, get_results, and recommend_next — that handle the full subject line A/B test lifecycle. Because marginal is hosted, there's nothing to deploy and no self-hosted server to maintain.
Connect any MCP-capable client (Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex) to the endpoint and you're ready to compare subject lines in one send.
- An MCP client pointed at https://marginal.sh/mcp
- A Bearer API key in the marg_live_... format
- Your recipient list and a draft email body
- Two or more candidate subject lines to compare
Step 1: Generate or supply your subject line variants
If you already have the lines you want to pit against each other, pass them directly to launch_test. If you want help, call generate_variants with your email context and let the agent draft several angles — curiosity, urgency, value-forward — for the same campaign.
Keep the variant count small so each line gets enough volume to produce a meaningful result within a single send.
- Prompt your agent: "Generate 3 subject line variants for this announcement."
- Review the output and trim weak or near-duplicate lines
- Aim for 2–3 distinct variants per test
Step 2: Launch the test in one send
Use launch_test to split your audience and dispatch every variant inside the same managed send. marginal handles the random assignment, delivery, and open/click tracking, so each subject line reaches a comparable slice of your list at the same time.
Sending everything together is what makes the comparison fair — timing, day-of-week, and list freshness are held constant across variants, isolating the subject line as the only variable.
- Call launch_test with your variants, body, and recipient list
- marginal allocates audience portions per variant automatically
- Tracking for opens and clicks is enabled by default
Step 3: Compare results and decide the winner
Once the send is out, call get_results to pull within-test lift metrics that show how each subject line performed against the others. You'll see open and click figures side by side, so the comparison is built into the data rather than something you reconstruct later.
To close the loop, run recommend_next — the agent uses the results to suggest a follow-up direction, whether that's a winning line to scale or a fresh angle to test. With the free tier offering 100 experiments per month, this marginal MCP tutorial loop is repeatable at no cost while you dial in your AI agent email marketing workflow.
- get_results returns per-variant opens, clicks, and lift
- Identify the top subject line by within-test performance
- Use recommend_next to plan the following experiment
- Read more 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