Testing Review Request Subject Lines in Claude Desktop
Asking customers for a review is high-stakes — the subject line decides whether your email gets opened at all. With marginal connected to Claude Desktop, you can draft, test, and optimize review request subject lines through plain conversation, no dashboard required.
Why subject lines make or break review requests
Review request emails compete with everything else in a busy inbox, and most go unopened. The framing of the subject line — gratitude, urgency, brevity, or a direct ask — has an outsized effect on open rates and ultimately the number of reviews you collect.
Rather than guessing which angle works for your audience, a subject line A/B test gives you evidence. marginal handles the variant generation, the split, the send, and the open/click tracking so you can focus on the wording.
- Different audiences respond to different tones — politeness vs. incentive vs. scarcity
- Small lift on a recurring review email compounds across every send
- Open-rate data tells you which framing earns attention before you scale it
Generating review request variants in Claude Desktop
Once marginal is connected as an MCP server, Claude Desktop can call generate_variants to produce a set of review request subject lines tuned to your product and tone. Describe the context — what the customer just bought, how long ago, whether you offer an incentive — and Claude proposes distinct candidates rather than minor rewordings.
From there, the conversation stays natural. You can ask Claude to make a variant warmer, shorter, or more direct, and the email MCP server keeps the candidate set ready to launch.
- "Generate 4 review request subject lines for a customer who bought running shoes two weeks ago"
- "Make one variant emphasize a thank-you and another emphasize speed"
- "Drop the emoji versions and keep them under 45 characters"
Launching and reading the subject line test
When the variants look right, ask Claude to launch_test and marginal handles the managed send — splitting your list, dispatching each variant, and tracking opens and clicks. There are no separate ESP credentials to wire up; marginal is hosted and runs the experiment end to end.
As results arrive, get_results reports within-test lift so you can compare variants on equal footing, and recommend_next suggests the winning subject line to roll forward. This closes the loop on AI agent email marketing: draft, test, measure, and decide all inside Claude Desktop.
- launch_test — split and send your review request variants
- get_results — open/click rates with within-test lift metrics
- recommend_next — surface the strongest subject line to scale
Connecting marginal to Claude Desktop
marginal exposes its tools over the standard MCP endpoint at https://marginal.sh/mcp, authenticated with a Bearer API key (marg_live_...). Add it to Claude Desktop's MCP configuration and the generate_variants, launch_test, get_results, and recommend_next tools become available in your chats.
The free tier covers 100 experiments per month, which is plenty to test review request subject lines across several customer segments. See the setup details and tool reference at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp · Registry: sh.marginal/mcp
- Auth: Bearer API key beginning with marg_live_
- Free tier: 100 experiments/month — no self-hosting required
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