marginal: a Customer.io Alternative for AI Agents
If you're wiring email into agent workflows, you don't need a full ESP dashboard — you need tools your model can call. marginal is an MCP-native Customer.io alternative for AI agents that turns experimentation, sends, and tracking into MCP calls your assistant can make directly.
Why an MCP email server instead of an ESP
Customer.io is built for marketers managing segments, journeys, and campaign UIs. That's a lot of surface area when your real goal is letting an AI agent draft, test, and send email. marginal exposes a focused email MCP server so the agent itself does the work — no glue code translating between an ESP REST API and your model's tool calls.
Because marginal speaks the Model Context Protocol natively, AI agent email marketing becomes a sequence of typed tool invocations instead of brittle prompt-to-API plumbing. The agent reasons, calls a tool, reads structured results, and decides what to do next.
- Customer.io: journeys, segments, and a campaign UI for humans
- marginal: generate_variants, launch_test, get_results, recommend_next for agents
- One endpoint (https://marginal.sh/mcp) authenticated with a Bearer API key
- Hosted — there's no self-hosted MCP server to operate
What you get for AI agent email
marginal covers the experimentation loop that matters most for AI agent email: writing candidate copy, running a subject line A/B test, sending, and measuring. The tracking and lift metrics come back as data the agent can act on, not charts a person has to interpret.
- generate_variants — produce subject line and body candidates to test
- launch_test — run a managed subject line A/B test and send it
- get_results — pull open/click tracking and within-test lift metrics
- recommend_next — let the agent decide the next experiment
- Free tier of 100 experiments/month to prototype agent workflows
Migrating the email step in your agent
If your agent currently calls Customer.io for transactional or marketing sends, the switch is mostly about replacing API calls with MCP tools. Add marginal to your MCP client, point it at the registry entry sh.marginal/mcp, and let the model handle variant generation and testing where it previously hit an ESP endpoint.
You keep the parts of your stack that work and swap only the email experimentation layer. The agent gains direct access to A/B testing and results without you maintaining a Customer.io integration on the side.
- Identify where the agent triggers email today
- Connect marginal as an MCP server in your client
- Replace ESP send/test calls with marginal tool calls
- Use get_results to feed decisions back into the agent loop
Works with the agent clients you already use
marginal connects to the MCP clients developers and growth engineers actually run, so an email MCP server fits into your existing setup rather than forcing a new one. Read the setup docs at https://marginal.sh/docs/ to get the endpoint and key configured.
- Cursor, Claude Code, Claude Desktop, and OpenAI Codex
- Windsurf, Cline, Continue, and Zed
- Single hosted endpoint: https://marginal.sh/mcp
- Bearer auth with a marg_live_ API key
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