Reading Experiment Results Through an Email MCP
When an AI agent launches a subject line A/B test, the next question is always the same: which variant won? marginal exposes experiment results over MCP so your agent can fetch open rates, click rates, and within-test lift without leaving the chat or copy-pasting from a dashboard.
Why fetch experiment results over MCP
Most marketers check test outcomes by logging into a separate tool. An email MCP server collapses that loop: the same agent that wrote the variants and started the send can also report the winner. That keeps the analysis grounded in the actual experiment your agent already has context on.
marginal is hosted, so there's no infrastructure to run. Point your client at https://marginal.sh/mcp, authenticate with a Bearer API key, and the get_results tool returns structured metrics your agent can reason over.
- No dashboard switching — results land in the agent conversation
- Structured JSON output your agent can summarize or chart
- Within-test lift metrics comparing each variant to the control
The tools behind the results
Experiment results don't appear in isolation — they're the payoff of a short workflow. marginal exposes four MCP tools that an AI agent calls in sequence to run a subject line A/B test end to end.
- generate_variants — draft competing subject lines for the test
- launch_test — start a managed send across your variants
- get_results — pull open/click tracking and within-test lift
- recommend_next — suggest the follow-up experiment based on what won
What get_results returns
Calling get_results gives your agent per-variant performance for an experiment results email MCP query: opens, clicks, and the lift each variant showed against the others within the same test. Because tracking is managed on marginal's side, you get consistent numbers without wiring up your own open and click pixels.
From there, a single prompt can have your agent interpret the data — for example, 'summarize the winner and propose three new subject lines that build on it' — which chains naturally into recommend_next and generate_variants.
- Open rate and click rate per subject line variant
- Within-test lift to identify a statistically meaningful winner
- Inputs for recommend_next to plan the following round
Getting started with AI agent email marketing
marginal works with Cursor, Claude Desktop, Claude Code, OpenAI Codex, Windsurf, Cline, Continue, and Zed. Add the endpoint and your marg_live_ key, then ask your agent to launch a subject line A/B test and report back when results are in.
The free tier covers 100 experiments per month, which is plenty to validate the results workflow before scaling. See the docs at https://marginal.sh/docs/ for tool schemas and example calls.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...)
- Registry: sh.marginal/mcp
- Docs: 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