How to Run a Referral Invite Subject Line Test
Referral invites live or die on the subject line. If the recipient doesn't open, the reward, the friend link, and the whole loop never start. This guide shows how to run a referral invite subject line test with marginal — a hosted email MCP server that lets AI agents generate variants, launch the test, and measure lift.
Why referral invite subject lines need their own test
Referral emails are a distinct genre. They're sent on behalf of a person, they carry a social cue ('your friend wants you to try this'), and they often promise a two-sided reward. That makes the subject line a balancing act between trust, curiosity, and the offer itself.
A dedicated referral invite subject line test isolates which of those levers matters most for your audience instead of guessing. Small wording shifts — naming the referrer, leading with the reward, or asking a question — can swing open rates meaningfully.
- Personal framing: 'Sara thinks you'll like this' vs. a brand-voice line
- Reward-forward: lead with the credit or discount both sides earn
- Curiosity vs. clarity: tease the offer or state it plainly
- Urgency: time-limited invite codes vs. evergreen links
Running the subject line A/B test with marginal
marginal exposes four MCP tools that map cleanly to the test workflow. An AI agent connected to the email MCP server can drive the whole loop from a single prompt, so you describe the referral campaign once and let the agent handle variant creation, send, and analysis.
Because marginal is hosted at https://marginal.sh/mcp, there's nothing to deploy. You authenticate with a Bearer API key and the managed sends, open tracking, and click tracking are handled for you.
- generate_variants — produce referral-flavored subject line candidates
- launch_test — start the subject line A/B test against your list segment
- get_results — pull open/click and within-test lift metrics
- recommend_next — let the agent suggest the winning line or a follow-up test
Letting an AI agent own the experiment
The point of AI agent email marketing is to compress the loop. Instead of copying subject lines into a dashboard, your agent in Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex calls the marginal tools directly during your normal workflow.
A typical referral invite subject line test prompt looks like: 'Generate five referral invite subject lines for our $20 two-sided credit, launch an A/B test to the dormant-user segment, and report lift after 24 hours.' The agent runs each step and surfaces the result.
- Keep variants tightly scoped to one hypothesis per test
- Compare within-test lift rather than absolute open rates across campaigns
- Use recommend_next to chain the winner into a follow-up reminder send
Getting started and free experiments
The free tier includes 100 experiments per month, which is plenty to test referral subject lines across several audience segments before committing budget to a winner. Each experiment gives you open and click tracking plus the lift comparison you need to call it.
Read the tool reference and authentication details in the docs at https://marginal.sh/docs/ to wire your client to the marginal email MCP server.
- Endpoint: https://marginal.sh/mcp with a marg_live_ Bearer key
- Registry: sh.marginal/mcp
- Free tier: 100 experiments/month
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