Referral Program Subject Lines in Claude Desktop
Connect Claude Desktop to marginal and run referral program subject line tests without leaving your chat window. Generate variants, launch a managed send, and let the data tell you which invitation copy earns more opens and shares.
Why test referral subject lines with an AI agent
Referral emails live or die on the subject line. "Give $10, get $10" lands very differently from "Your friends will thank you for this." Tone, incentive framing, and urgency all shift open rates, and the only honest way to know what works for your list is a subject line A/B test.
With marginal connected to Claude Desktop, you describe the campaign in plain language and the agent handles variant generation, the send, and the results. That makes AI agent email marketing practical for a referral program where you want to iterate quickly across multiple incentive angles.
- Test incentive framing: dollar amount vs. percentage vs. perks
- Compare warm-referral tone against transactional reward language
- Measure open and click lift before committing your whole list
- Iterate weekly as your referral offer evolves
Connecting Claude Desktop to marginal
marginal is a hosted email MCP server, so there is nothing to deploy. Add the endpoint to your Claude Desktop MCP configuration with a Bearer API key from your marginal account, and the four tools — generate_variants, launch_test, get_results, and recommend_next — become available to the agent.
The free tier covers 100 experiments per month, which is plenty for refining a referral program subject line through several rounds of testing. See the docs at https://marginal.sh/docs/ for the full configuration reference.
- Registry: sh.marginal/mcp
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...)
- Works in Claude Desktop, Claude Code, Cursor, Windsurf, and more
Running the referral subject line test
A typical workflow inside Claude Desktop starts with a prompt like: "Generate five subject lines for our refer-a-friend launch where both people get $15, then launch an A/B test." The agent calls generate_variants to draft the options, launch_test to send them, and tracks opens and clicks automatically.
Once enough data accrues, ask Claude Desktop to pull get_results. marginal reports within-test lift so you can see which referral subject line is genuinely outperforming, not just which got lucky early. Then recommend_next suggests the strongest follow-up direction.
- generate_variants — draft multiple referral subject lines
- launch_test — run the managed subject line A/B test
- get_results — review open/click rates and lift metrics
- recommend_next — get the next variant to try
Getting more from each test round
Treat each referral campaign as a learning loop. Keep your audience and timing consistent so the subject line is the variable, and feed winning angles back into the next generate_variants prompt. Over a few cycles, Claude Desktop and marginal help you converge on the incentive language your subscribers actually respond to.
Because everything runs through the agent, you can document what won and why directly in your chat history — handy when you hand the referral program off to another teammate or revisit it next quarter.
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