The launch_test MCP Tool: Start Subject Line Tests From Your Agent
launch_test is the marginal MCP tool that turns a set of subject line variants into a live A/B test. It schedules the managed send, splits your audience, and begins tracking opens and clicks — all from a single tool call your AI agent makes.
What launch_test does
Within marginal's email MCP server, the workflow is deliberately small: generate candidates, launch the test, read results, decide next. The launch_test MCP tool sits at the second step. Once you have variants (often produced by generate_variants), launch_test commits them to a real experiment and starts the send.
- Accepts your subject line variants and the test configuration
- Splits the audience and runs a subject line A/B test
- Triggers the managed send so you never touch raw SMTP
- Begins open and click tracking automatically
- Returns a test identifier you later pass to get_results
Where it fits in the AI agent email marketing loop
For AI agent email marketing, launch_test is the action step — the moment intent becomes a live experiment. An agent in Cursor, Claude Code, or Windsurf can call generate_variants to draft subject lines, then immediately call launch_test to put them in front of real recipients.
After the test runs, the agent calls get_results for within-test lift metrics and recommend_next for a data-backed suggestion. Because every tool shares the same hosted backend, your agent doesn't juggle send infrastructure or tracking pixels.
- generate_variants → draft subject lines
- launch_test → start the managed send and tracking
- get_results → opens, clicks, and lift
- recommend_next → what to test or send next
Connecting and authenticating
marginal is hosted, so there's nothing to deploy. Point your MCP client at the endpoint https://marginal.sh/mcp and authenticate with a Bearer API key in the marg_live_ format. Once connected, launch_test appears alongside the other tools in your client's tool list.
The free tier covers 100 experiments per month, which is plenty for iterating on subject line tests while you wire up an agent workflow. See the configuration details and tool schemas at https://marginal.sh/mcp and the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Registry: sh.marginal/mcp
- Auth: Bearer key (marg_live_...)
- Works with Cursor, Claude Desktop, Codex, Cline, Continue, and Zed
Tips for reliable subject line tests
Treat each launch_test call as a single, well-scoped experiment. Keep variants meaningfully different so the lift signal is interpretable, and let the test gather enough opens before acting on results.
- Test 2–4 distinct subject lines per experiment, not minor tweaks
- Wait for get_results before declaring a winner
- Use recommend_next to compound learnings across sends
- Stay within your monthly experiment budget on the free tier
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