Subject Line A/B Testing in Claude Desktop
Connect Claude Desktop to marginal's email MCP server and run a full subject line A/B test as a conversation — generate variants, send to a sample, and let the agent recommend the winner.
Why run A/B tests from Claude Desktop
Claude Desktop already lets you draft and iterate on copy in plain language. With marginal connected as an MCP server, that same chat window becomes a launchpad for live subject line A/B tests — no separate dashboard, no copy-pasting between tools.
marginal is a hosted email MCP server, so there's nothing to deploy. You add the endpoint once, paste your Bearer key, and Claude Desktop can call the experiment tools directly.
- Ask Claude to brainstorm five subject lines, then test the top two in one session
- Keep the whole subject line A/B test workflow inside a single conversation
- Free tier covers 100 experiments per month to get started
Connect marginal to Claude Desktop
Add marginal's MCP endpoint to your Claude Desktop config so the agent can reach the experiment tools. The connection is authenticated with a Bearer API key in the marg_live_ format from your marginal account.
- Endpoint: https://marginal.sh/mcp
- Registry name: sh.marginal/mcp
- Auth: Bearer token (marg_live_...)
- Full setup steps live at https://marginal.sh/mcp and https://marginal.sh/docs/
The Claude Desktop agent workflow
A typical subject line A/B test in Claude Desktop moves through four marginal tools in order. You describe the audience and goal, and the agent chains the calls — generating candidates, launching the split, then reading results once opens and clicks come in.
Because marginal handles the managed sends and tracking, Claude Desktop focuses on the decisions: which variants to test and what to do next.
- generate_variants — produce subject line candidates from a prompt
- launch_test — start the A/B split and managed send
- get_results — pull open/click and within-test lift metrics
- recommend_next — let the agent suggest the winning line or a follow-up
Reading lift and picking a winner
After a test runs, ask Claude Desktop to summarize the results. marginal reports open and click tracking plus within-test lift, so the agent can tell you which subject line outperformed the control and by how much.
This makes AI agent email marketing practical: instead of eyeballing percentages, you get a clear recommendation grounded in the test data, ready to roll into your next send.
- Compare open and click rates per variant
- Use within-test lift to judge significance of the difference
- Call recommend_next to plan the follow-up campaign
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