Product-Led Growth Email MCP for AI Agents
Product-led growth teams live and die by activation, onboarding, and re-engagement emails. marginal is a hosted email MCP server that lets your AI agents generate, test, and send those emails — and read back which subject line actually won.
Why PLG teams need an email MCP server
In a product-led motion, lifecycle email is part of the product surface: trial nudges, feature adoption prompts, and win-back flows all drive the metrics you care about. The bottleneck is rarely ideas — it's the loop of writing variants, running a clean subject line A/B test, and acting on the results.
A product led growth email MCP closes that loop inside the tools you already use. Instead of switching to a marketing dashboard, your agent calls marginal directly to spin up experiments and pull lift data, so iteration happens at the speed of a chat prompt.
- Generate onboarding and activation copy as variants, not one-off drafts
- Run controlled subject line A/B tests with measured within-test lift
- Let an AI agent decide the next experiment from real open/click data
- Keep the whole loop in Cursor, Claude, or your editor of choice
What the marginal MCP tools do
marginal exposes four tools over the Model Context Protocol so an AI agent can run an entire experiment without leaving the conversation. The endpoint is https://marginal.sh/mcp and auth is a single Bearer API key (marg_live_...) — there's nothing to self-host.
- generate_variants — produce subject line options for a PLG email
- launch_test — start a managed subject line A/B test and send it
- get_results — read open/click tracking and within-test lift metrics
- recommend_next — get a data-backed suggestion for the next test
A PLG experiment loop with AI agent email marketing
AI agent email marketing works best when each step feeds the next. With marginal, you describe the goal — say, lifting day-3 activation opens — and the agent generates variants, launches the test, waits for tracking data, then recommends what to try next.
Because lift is measured within each test, you're comparing variants under the same conditions rather than guessing from raw open rates across different sends. That makes results easy for an agent to reason about and act on.
- Prompt: "Test three subject lines for our trial-expiry email"
- Agent calls generate_variants, then launch_test against your list
- get_results returns open/click rates and the winning variant's lift
- recommend_next proposes the follow-up experiment to keep iterating
Getting connected
marginal is a hosted email MCP server, so connecting is a matter of adding the endpoint and your API key to any MCP-compatible client — Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex. The free tier covers 100 experiments per month, which is plenty to validate a PLG email program before scaling.
Full setup steps and tool references live in the docs at https://marginal.sh/docs/, and the MCP endpoint details are at https://marginal.sh/mcp.
- Hosted — no infrastructure to run or maintain
- Works across every major MCP client
- 100 experiments/month on the free tier
- Registry name: sh.marginal/mcp
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