marginal: a HubSpot Alternative for AI Agents
If your workflow starts in Cursor or Claude Code rather than a marketing dashboard, HubSpot's UI-first model gets in the way. marginal is a hosted email MCP server built so AI agents can generate, launch, and measure email tests without leaving the conversation.
Where HubSpot and marginal diverge
HubSpot is a full marketing and CRM suite designed for human marketers clicking through screens. That breadth is useful for some teams, but it means automating email tests usually involves stitching together REST endpoints, workflow builders, and webhook plumbing.
marginal takes the opposite approach. It exposes a small set of MCP tools that an agent calls directly, so AI agent email marketing happens through structured tool calls instead of a browser session. There is no dashboard to navigate and nothing to self-host.
- HubSpot: broad CRM + marketing platform, UI-centric, manual campaign setup.
- marginal: focused MCP email server for AI agent email, agent-driven by design.
- HubSpot fits human marketing teams; marginal fits agents and growth engineers.
- marginal is hosted at https://marginal.sh/mcp — no infrastructure to run.
Built for subject line A/B tests via MCP
The core job marginal does well is the subject line A/B test. An agent asks for variants, launches a test against your list, and reads back which line is winning — all as MCP tool calls.
Four tools cover the loop: generate_variants drafts subject options, launch_test sends them as a managed experiment, get_results returns open and click tracking with within-test lift metrics, and recommend_next suggests the follow-up move.
- generate_variants — produce candidate subject lines for a test.
- launch_test — run a managed send with open/click tracking.
- get_results — pull within-test lift and engagement numbers.
- recommend_next — let the agent decide the next experiment.
Connect from the tools you already use
marginal speaks MCP, so it plugs into the editors and agent clients you already run. Add the endpoint, authenticate with a Bearer API key, and your agent can run email experiments inline.
Supported clients include Cursor, OpenAI Codex, Claude Desktop, Claude Code, Windsurf, Cline, Continue, and Zed. Authentication uses an API key in the form marg_live_..., and the endpoint lives at https://marginal.sh/mcp.
- Registry: sh.marginal/mcp · Endpoint: https://marginal.sh/mcp
- Auth: Bearer API key (marg_live_...).
- Works across major MCP clients — no plugin marketplace required.
- Docs and setup details: https://marginal.sh/docs/
When to choose marginal over HubSpot
Pick a HubSpot alternative for AI agents when your goal is iterating on subject lines and measured sends programmatically, not managing contacts, pipelines, and landing pages. marginal does one slice of the email problem and exposes it cleanly to agents.
You can validate the workflow on the free tier of 100 experiments per month before committing, which is enough to wire an agent's test-and-learn loop end to end.
- Choose marginal for agent-native subject line A/B tests and managed sends.
- Stay with HubSpot if you need a full CRM and human-run campaigns.
- Free tier: 100 experiments/month to prototype AI agent email.
- marginal complements, rather than replaces, a full marketing stack.
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