Testing Review Request Subject Lines in OpenAI Codex
Pair OpenAI Codex with marginal to draft, launch, and measure review request subject lines without leaving your editor. This guide walks through a focused subject line A/B test for post-purchase review emails using the email MCP server.
Why review request emails need subject line testing
Review request emails ask a customer to do something they often forget: leave a rating after a purchase. The subject line is the only thing standing between your message and the trash, so small wording changes can swing open rates meaningfully.
Common angles worth testing include gratitude, urgency, product specificity, and incentive mentions. Rather than guessing, a subject line test lets the data tell you which framing earns the open for your audience.
- "How was your order?" vs "Got 30 seconds to rate it?"
- Including the product name vs a generic ask
- Personalization with first name vs none
- Plain ask vs incentive-led ("Leave a review, get 10% off")
Connecting marginal to OpenAI Codex
marginal is a hosted email MCP server, so OpenAI Codex talks to it over the endpoint at https://marginal.sh/mcp with a Bearer API key. There is nothing to self-host — once Codex can reach the server, the four tools become available as agent actions.
The tools you'll use for a review request subject line test are generate_variants to brainstorm, launch_test to send the experiment, get_results to read open and click data, and recommend_next to decide the follow-up.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key in the form marg_live_...
- Registry id: sh.marginal/mcp
- Free tier covers 100 experiments per month
Running the review request subject line test
Inside OpenAI Codex, describe the campaign in plain language and let the AI agent email marketing flow do the rest. Ask Codex to generate review request subject lines aimed at recent buyers, then launch the variants as a split test against your list.
marginal handles the managed send and tracks opens and clicks per variant, reporting within-test lift so you can see which subject line actually moved the needle rather than relying on raw numbers alone.
- Prompt: "Generate 5 review request subject lines for customers who bought last week."
- Run launch_test to split traffic across the variants
- Poll get_results for opens, clicks, and lift metrics
- Call recommend_next to pick the winner or queue a second round
Reading results and iterating
Once enough sends accumulate, get_results surfaces the leading subject line along with confidence in the lift. For review request campaigns, prioritize open rate as the primary signal since the click happens after the open.
Use recommend_next to turn findings into the next experiment automatically. A first test might reveal that specificity wins; a follow-up subject line A/B test can then explore incentive language built on that winning frame. See https://marginal.sh/docs/ for tool details.
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