Running a Merge Tag Subject Line Experiment
Personalized subject lines that pull in a first name, city, or recent product can lift opens — but only if you test them properly. Here's how to run a merge tag subject line experiment through marginal, the hosted email MCP server, so your AI agent can generate, launch, and measure personalized variants without leaving your editor.
Why test merge tags in the subject line
Merge tags like {{first_name}} or {{last_order}} promise relevance, but they don't always win. A name in the subject can feel warm in one audience and intrusive in another, and a fallback that renders awkwardly can sink your open rate. The only reliable way to know is a controlled subject line A/B test that compares the personalized variant against a clean static one.
A merge tag subject line experiment isolates that single variable so you can attribute any change in opens to personalization itself — not to a different hook, length, or send time.
- Compare {{first_name}} vs. no name on the same offer
- Test fallback copy when the merge field is empty
- Check whether deeper tags (city, product) beat a simple name
- Measure within-test lift instead of guessing
How marginal handles it
marginal exposes four MCP tools your AI agent calls directly: generate_variants drafts personalized and control subject lines, launch_test splits your list and sends them, get_results returns open and click tracking, and recommend_next suggests the follow-up based on observed lift. Because it's a hosted email MCP server, there's nothing to self-host — you authenticate with a Bearer API key and the endpoint at https://marginal.sh/mcp does the sends and tracking.
For a merge tag experiment specifically, you describe the personalization you want and the agent produces matched variants where the only difference is the merge tag, keeping the test honest.
- generate_variants — personalized and control subject lines with fallbacks
- launch_test — managed split send to your audience
- get_results — open/click tracking and within-test lift
- recommend_next — data-driven next step
A typical workflow
From an AI agent in Cursor, Claude Code, or Windsurf, the flow is conversational. You ask the agent to build the experiment, and it chains the MCP tools for you while you review each step.
- Ask: "Create a subject line A/B test with one {{first_name}} variant and one plain control."
- The agent calls generate_variants and shows both options plus the empty-name fallback.
- Approve, and launch_test splits and sends through marginal.
- Pull get_results to see which variant earned more opens and the lift between them.
- Let recommend_next propose whether to roll out personalization or test a deeper merge tag.
Getting started
marginal is built for AI agent email marketing, so the entire merge tag subject line experiment runs from natural-language prompts inside your editor. The free tier covers 100 experiments per month — plenty for iterating on personalization strategy before you commit to a winner.
Add your API key, point your client at https://marginal.sh/mcp, and see the docs at https://marginal.sh/docs/ for tool schemas and merge tag formatting.
- Works with Cursor, Claude Desktop, Claude Code, Codex, Windsurf, Cline, Continue, and Zed
- Hosted — no infrastructure to run
- 100 experiments/month on the free tier
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