Managed Email Sends for Agents
When an AI agent decides what to send, it shouldn't also have to babysit SMTP, suppression lists, or test bookkeeping. marginal gives agents managed email sends — the actual delivery and measurement happen behind a hosted MCP server, so your agent just asks for an outcome and reads the results.
What 'managed' means here
Managed email sends for agents means the send pipeline is handled for you. Instead of wiring an agent to a raw email provider, you connect it to marginal's email MCP server and call a small set of tools. The server schedules the send, tracks opens and clicks, and computes within-test lift so the agent gets numbers it can reason about — not a delivery problem to solve.
- generate_variants — produce subject line candidates for a campaign
- launch_test — start a subject line A/B test and dispatch the send
- get_results — pull open/click rates and within-test lift
- recommend_next — get a data-backed suggestion for the next move
Why agents need this instead of direct API calls
An AI agent doing email marketing on its own tends to hallucinate the boring parts: how a split was allocated, which variant actually won, whether the difference was meaningful. By routing through marginal, the agent works against a consistent contract. The subject line A/B test setup, randomization, and measurement are owned by the server, so results are reproducible across runs and across different LLMs.
- No need to manage send infrastructure inside the agent
- Consistent lift metrics the agent can compare across tests
- Open and click tracking included with every managed send
- Stable tool surface across Claude, Cursor, Codex, and more
A typical agent loop
Most AI agent email marketing workflows fall into a simple loop: draft variants, launch a test, wait for signal, then act on the recommendation. marginal supports that loop end to end so the agent can run it autonomously or with a human approving each step.
Because the sends are managed, the agent can launch a test in one turn and return later to read results without holding any state itself. That makes it practical to embed inside scheduled jobs, chat sessions, or longer-running planning agents.
- Call generate_variants with your offer and audience context
- Call launch_test to dispatch and split traffic across subjects
- Poll get_results for opens, clicks, and lift
- Call recommend_next to choose the winning direction
Getting connected
marginal is fully hosted — there's no self-hosted MCP server to deploy. Point any MCP-capable client at https://marginal.sh/mcp, authenticate with a Bearer API key (marg_live_...), and your agent has managed email sends available as tools. The free tier covers 100 experiments per month, which is enough to validate a real subject line testing workflow before scaling up.
Full tool schemas and setup steps live in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Registry name: sh.marginal/mcp
- Works with Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, and Codex
- Free tier: 100 experiments/month
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