Email Attribution for Subject Lines
Knowing which subject line actually moved the needle is harder than it looks. Email attribution for subject lines means tying opens and clicks back to the exact variant a recipient received — and doing it inside a controlled test so the numbers mean something. marginal handles that attribution for you through an email MCP server your AI agent can call directly.
Why subject line attribution is tricky
If you blast two subject lines on different days, or to different segments, you can't cleanly credit a lift to the subject line itself. Send time, audience composition, and content all leak into the result. Proper attribution requires that variants are assigned within a single randomized test so the only thing that differs is the line a recipient sees.
marginal frames every subject line A/B test as a within-test comparison. Recipients are split across variants, opens and clicks are tracked per variant, and lift is measured relative to a shared baseline — so the attribution holds up.
- Randomized assignment removes audience and timing bias
- Open and click tracking is recorded at the variant level
- Within-test lift metrics show how each line performed against the others
How marginal attributes results
marginal is a hosted email MCP server. Your AI agent generates candidate lines, launches a test, and pulls back attributed results — all through a small set of MCP tools. There's no spreadsheet reconciliation and no manual UTM bookkeeping for the subject line dimension.
- generate_variants — produce subject line candidates to test
- launch_test — split your audience and start a managed send
- get_results — read per-variant opens, clicks, and lift
- recommend_next — get a data-backed suggestion for the next iteration
Running an attributed test from an AI agent
This is where AI agent email marketing gets practical. Connect marginal to your agent in Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex, then let the agent drive the loop: draft variants, launch, wait for results, and read attribution without you switching tools.
Because attribution is computed inside the test, the agent receives clean signal it can reason about — which line earned the higher open rate, and by how much — rather than raw counts it has to interpret.
- Point your MCP client at https://marginal.sh/mcp
- Authenticate with a Bearer API key (marg_live_...)
- Let the agent iterate on subject lines using attributed lift
Getting started
The free tier covers 100 experiments per month, which is plenty to validate that your attribution loop works before scaling up. See the docs at https://marginal.sh/docs/ for tool schemas and example calls, and connect your client to the endpoint at https://marginal.sh/mcp to start measuring which subject lines actually win.
- Free tier: 100 experiments/month
- Registry: sh.marginal/mcp
- Docs: 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