building
Iterable MCP Server
“How many campaigns did we send last week?”
Ask an AI assistant that question with Iterable’s MCP server plugged in and you get a number back. No dashboard opened, not a line of code written. Claude, Cursor, Claude Code, Gemini CLI, and a few others can all do it now. This is Iterable’s official MCP server, and it is what lets them talk to your Iterable account in plain English.
The thing doing the talking is a TypeScript bridge between the Model Context Protocol and Iterable’s API. Through it an assistant can pull campaigns, templates, journeys, and user data. It can create and manage those resources. It can send across email, SMS, push, in-app, and WhatsApp.
Read that last sentence again. An agent that can count last week’s sends is also an agent that can send. To real people, from your account. Wiring a model to a live account is the easy part. The work is deciding what it is allowed to touch, and making it ask first.
So the server ships with a three-tier permission model: read, write, send. Credentials go in the OS keychain instead of a dotfile. And before it can touch PII or fire a real message, you have to opt in. Most of the effort in this project went into that permission model, not the plumbing.
I helped build it. If Nova is the agent, this is one
of the doors it knocks on. MIT licensed, and npx @iterable/mcp setup if you
want to try it.