A SendGrid Alternative Built for Experiments
If you reach for SendGrid to send transactional or marketing email but find yourself bolting on spreadsheets and custom scripts just to run a subject line A/B test, you want a tool shaped around experiments instead of raw delivery. marginal is a hosted email MCP server that lets AI agents generate, launch, and measure tests directly.
Why look past SendGrid for experimentation
SendGrid is a solid delivery platform, but it treats experimentation as an afterthought. Splitting an audience, tracking lift, and deciding a winner usually means stitching together its API, webhooks, and your own analytics. When your goal is iterating on copy fast, that overhead slows you down.
A sendgrid alternative for experiments should make the test the primary object — not the send. marginal does exactly that: every action centers on variants, audiences, and measured outcomes.
- Experiment-first design instead of delivery-first plumbing
- Built-in open/click tracking with within-test lift metrics
- No glue code to compare variant performance
- Free tier of 100 experiments per month to evaluate it
An email MCP server your AI agent can drive
marginal exposes its functionality as an email MCP server, so AI agent email marketing workflows can call it natively. Connect a client like Cursor, Claude Code, Windsurf, or Cline and your agent gains four tools: generate_variants, launch_test, get_results, and recommend_next.
Instead of writing imperative SendGrid API calls, you describe the outcome you want and the agent orchestrates the experiment through https://marginal.sh/mcp.
- generate_variants — draft multiple subject line candidates
- launch_test — split your audience and send the managed test
- get_results — pull opens, clicks, and lift per variant
- recommend_next — get a data-backed suggestion for the next round
Running a subject line A/B test with marginal
A typical subject line A/B test flow takes minutes. Ask your agent to generate variants for a campaign, launch the test against your list, and let marginal handle the managed send and tracking. When results arrive, lift metrics show which subject line won inside the test, and recommend_next points you toward the following iteration.
Because the whole loop lives behind the MCP tools, you can keep experimenting without leaving your editor or copying numbers between tools.
Getting connected
marginal is hosted, so there is nothing to deploy or self-host. Add the endpoint https://marginal.sh/mcp to your MCP client, authenticate with a Bearer API key in the marg_live_ format, and start launching tests.
It works across Claude Desktop, OpenAI Codex, Continue, Zed, and the other supported clients. See the docs at https://marginal.sh/docs/ for client-specific setup.
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