marginal: a Mailchimp alternative built for AI agents
Mailchimp was designed for marketers clicking through a dashboard. marginal is designed for agents calling tools. If you're building AI agent email marketing and want experiments to run inside your workflow, here's how the MCP-native approach differs.
Why a Mailchimp alternative for AI agents looks different
Traditional ESPs like Mailchimp expose REST APIs that assume a human is orchestrating campaigns, reviewing reports, and deciding what to send next. An autonomous agent has to stitch those endpoints together, parse dashboards it can't see, and guess at what "good" looks like.
marginal flips the model. It's an email MCP server, so your AI agent talks to it through the Model Context Protocol with structured tools instead of brittle API glue. The agent generates variants, launches a test, and reads back lift metrics — no screen-scraping, no manual campaign wizard.
- MCP-native tools instead of dashboard-first REST flows
- Built around the experiment loop agents actually run
- Structured results an agent can reason over directly
What marginal covers (and what it doesn't)
marginal is focused: it does subject line A/B testing, managed sends, and open/click tracking with within-test lift metrics. That's the slice of Mailchimp most relevant to AI agent email — the part where you want fast iteration and clear winners.
It is not a full marketing suite. If you need landing pages, audience CRM, or visual automation builders, Mailchimp still does those things. marginal is the piece you point an agent at when the job is testing and sending email and learning from the result.
- generate_variants — draft subject line and copy options
- launch_test — start a managed subject line A/B test
- get_results — pull opens, clicks, and within-test lift
- recommend_next — get a data-backed next step
Hosted, with no MCP server to run
marginal is a hosted MCP email server. You don't deploy or maintain anything — point your client at the endpoint and authenticate with a Bearer API key (marg_live_...). The registry name is sh.marginal/mcp and the endpoint lives at https://marginal.sh/mcp.
It works across the agent clients teams already use: Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed. Setup details are in the docs at https://marginal.sh/docs/.
- Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key, no ESP credentials to manage in your code
- Free tier: 100 experiments per month to evaluate the workflow
A typical AI agent email marketing loop
Here's how the tools chain together in practice. An agent drafts a handful of subject lines, ships them as a subject line A/B test, waits for engagement data, then acts on the recommendation — all without a human opening a campaign editor.
Because every step returns structured data, the agent can decide on its own whether to roll out the winner or keep iterating, which is exactly the autonomy Mailchimp's human-centric flow makes awkward.
- Agent calls generate_variants to produce subject candidates
- launch_test runs the managed A/B split and sends
- get_results returns opens, clicks, and lift per variant
- recommend_next suggests the winning send or another round
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