Usage Limit for Email Subject Tests with marginal
If you're running a usage limit email subject test, you want to know exactly what counts against your quota before you wire an AI agent into your sends. Here's how marginal's free tier maps to real subject line A/B testing work.
What counts toward your usage limit
marginal is a hosted email MCP server that gives AI agents tools to draft, launch, and measure subject line tests. The free tier includes 100 experiments per month, and each launched test counts as one experiment regardless of how many subject variants it carries.
- Launching a subject line A/B test via launch_test = 1 experiment
- Generating draft variants with generate_variants does not consume a launch slot
- Pulling open/click data with get_results is free to call repeatedly
- recommend_next suggests follow-ups without auto-spending your quota
Planning a subject test within the limit
Because the usage limit on the free tier is counted per experiment, you have room to run a meaningful cadence of subject line A/B tests without immediately upgrading. The trick is batching your variants into a single test rather than firing separate launches for each idea.
An AI agent email marketing workflow benefits here: the agent can generate ten candidate subject lines, narrow them to two or three with reasoning, and then launch one test — keeping your experiment count lean while still exploring widely.
- Generate freely, launch deliberately
- Group competing subject lines into one test to conserve experiments
- Use within-test lift metrics to decide the winner before launching a sequel test
- Reserve experiment slots for high-traffic segments where lift matters most
How an AI agent uses the tools
Once your client is connected to the marginal email MCP server at https://marginal.sh/mcp, the agent calls four tools in sequence. The usage limit only bites at the launch step, so the exploratory parts of your workflow stay cheap.
This makes it practical to keep a subject-testing loop running across newsletters and lifecycle emails without surprise overages on the free tier.
- generate_variants — draft candidate subject lines
- launch_test — start the managed send and A/B split (counts as 1 experiment)
- get_results — track opens, clicks, and within-test lift
- recommend_next — plan the next subject test from observed data
When you hit the limit
If you reach 100 experiments in a month, your existing tests keep tracking and you can still read results — the limit applies to launching new ones. For teams that need more headroom, plan around the monthly reset or review options in the docs.
Connect any supported client — Cursor, Claude Desktop, Claude Code, Windsurf, Cline, Continue, Zed, or OpenAI Codex — and authenticate with your Bearer API key to start measuring subject lines against the same shared quota.
- Quota resets monthly on the free tier
- Read-only calls (results, recommendations) remain available
- See https://marginal.sh/docs/ for setup and tool details
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