How to Run Lifecycle Email Experiments with marginal
Lifecycle emails — welcome series, onboarding nudges, win-back flows — deserve the same experimentation rigor as paid acquisition. This how-to guide walks through running lifecycle email experiments end to end using marginal's MCP tools, so your AI agent can generate, launch, and learn from each send.
Connect the email MCP server
marginal is a hosted email MCP server, so there's nothing to deploy. Point your MCP client at the endpoint and authenticate with a Bearer API key. Any MCP-capable client works: Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex.
Once connected, your agent has access to four tools — generate_variants, launch_test, get_results, and recommend_next — which map cleanly onto the experiment loop you'll use for every lifecycle stage.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key in the form marg_live_...
- Registry id: sh.marginal/mcp
- Free tier covers 100 experiments/month — enough to cover most lifecycle flows
Step 1: Generate variants for the lifecycle stage
Start by deciding which lifecycle message you're testing — a day-one welcome, a trial-expiry reminder, or a dormant-user win-back. Ask your agent to call generate_variants with the message intent and audience context. This produces multiple subject line and copy options tailored to that stage.
Because lifecycle emails carry different goals (activation vs. reactivation vs. retention), feed the tool that intent so the variants are framed correctly rather than generic.
- Describe the lifecycle trigger and the desired action
- Generate 2–4 subject line variants per test to keep results readable
- Keep one variant close to your current baseline as a control
Step 2: Launch the test and track engagement
Use launch_test to run a subject line A/B test as a managed send. marginal handles delivery and tracks opens and clicks for you, so you don't wire up separate analytics. This is the core of how to run lifecycle email experiments without building infrastructure.
Split your lifecycle segment across the variants and let the test gather signal. Each variant accrues open and click data that feeds the within-test lift metrics.
- launch_test sends variants to your chosen segment
- Open and click tracking is built in
- Lift is measured within the test, so comparisons stay apples-to-apples
Step 3: Read results and decide the next move
Call get_results to pull within-test lift across your variants — which subject line earned more opens, which drove more clicks. For lifecycle flows, watch the click metric closely since the real goal is the next action, not just the open.
Then call recommend_next to let the AI agent suggest the follow-up experiment: promote the winning variant to the live flow, iterate on the runner-up, or test a new angle for the same stage. This is where AI agent email marketing turns a single test into a continuous improvement loop across your lifecycle program.
- get_results returns open/click and within-test lift
- recommend_next proposes the follow-up experiment
- Repeat the loop per lifecycle stage to compound gains
- See the marginal MCP tutorial and docs at 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