Preheader Text A/B Testing with the marginal MCP Server
Subject lines get most of the attention, but the preheader — that snippet of text rendered next to or below the subject in the inbox — is a second headline you control. A preheader text A/B test MCP lets your AI agent draft variants, send them, and measure which one earns more opens, all without leaving your editor.
Why test preheader text, not just subject lines
The preheader is the first sentence many subscribers read before deciding whether to open. When it merely repeats the subject or shows a stray 'View in browser' line, you waste prime inbox real estate. Testing it independently surfaces which framing — a benefit, a number, a question, a continuation of the subject — actually moves open rate.
Pairing a preheader test with a subject line A/B test gives you a clearer picture of inbox performance, since the two render together. marginal treats both as variant dimensions you can experiment on through the same workflow.
- Preheaders extend or contrast the subject's promise
- A weak preheader can drag down an otherwise strong subject
- Open lift from preheader changes is often underestimated
- Small, frequent tests build a library of what your list responds to
How an email MCP server runs the test
marginal is a hosted email MCP server, so the experiment loop lives behind four MCP tools your agent can call. You describe the email and audience in natural language, and the agent handles variant creation, the managed send, and reporting.
A typical preheader A/B test runs like this:
- generate_variants — draft multiple preheader options for the same subject
- launch_test — split your audience and send the variants
- get_results — pull open and click tracking with within-test lift metrics
- recommend_next — let the agent suggest the follow-up experiment
Set it up with your AI agent
Connect marginal to Cursor, Claude Desktop, Claude Code, OpenAI Codex, Windsurf, Cline, Continue, or Zed. The MCP endpoint is https://marginal.sh/mcp and authentication is a Bearer API key in the marg_live_ format — no Mailchimp or Klaviyo credentials required.
Once connected, AI agent email marketing becomes a conversation: ask the agent to test three preheaders against your best-performing subject, and it returns ranked results when the data is in. The free tier covers 100 experiments per month, which is plenty for steady preheader iteration.
- Registry: sh.marginal/mcp
- Endpoint: https://marginal.sh/mcp
- Docs and tool reference: https://marginal.sh/docs/
Reading lift and deciding the next move
After the send, get_results reports open and click rates per variant alongside within-test lift, so you can see how much the winning preheader outperformed the rest within the same audience split. Because the comparison happens inside one test, you avoid the noise of comparing across separate campaigns.
From there, recommend_next helps you compound the learning — keep the winning preheader and test a new subject, or hold the subject and probe a different preheader angle. Over time these paired preheader and subject line tests sharpen the instincts your agent applies to every future send.
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