createsend-mcp

A Model Context Protocol server that exposes every Campaign Monitor API operation (113 tools) over stdio. The tool table is generated from the Campaign Monitor OpenAPI spec; each tool is dispatched straight to the API over HTTP, so the server is self-contained with no SDK dependency.

Package managers

Examples below use npm. For pnpm or Yarn, swap commands as follows:

| Task | npm | pnpm | Yarn | | --- | --- | --- | --- | | Install dependencies | npm install | pnpm install | yarn | | Run a package script | npm run <name> | pnpm <name> | yarn <name> | | One-off run (no install) | npx @create-send/mcp | pnpm dlx @create-send/mcp | yarn dlx @create-send/mcp |

Run

CREATESEND_API_KEY=... npx @create-send/mcp

…or from source:

npm install
npm run generate
npm run mcp

Wire it into Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "createsend": {
      "command": "npx",
      "args": ["-y", "@create-send/mcp"],
      "env": { "CREATESEND_API_KEY": "your-key" }
    }
  }
}

For pnpm or Yarn, set "command" to "pnpm" / "yarn" and "args" to ["dlx", "@create-send/mcp"], keeping the same env.

Tool naming and inputs

Tool names match the spec's operationIds (e.g. getCampaignOpens, getListStats, addSubscribersConsentToTrack). Inputs follow the same shape across every tool:

• path params at the top level (e.g. id, listId),
• query string parameters under query,
• request body under body.

So a call to getCampaignOpens looks like:

{
  "id": "CAMPAIGN_ID",
  "query": { "date": "2026-01-01", "page": 1, "pageSize": 1000 }
}

On success the tool returns the API response body as text. API errors come back as the error payload with isError: true (the constructor still throws if the API key is missing).

Built-in guidance

The server ships guidance so any client can use it well without prior knowledge:

• Instructions — sent on connect and surfaced to the model automatically: tool shape, API-key/client scope, and the rules for getting HTML email in.
• Prompt import-html-email — a step-by-step for turning HTML email into a template + campaign and sending it (clients surface prompts as commands).
• Resource createsend://guides/html-email — a reference covering hosting, template validation errors, the create-then-send flow, key scope, and personalization.

These live in src/guidance.ts and are not regenerated by npm run generate.

import_html_template

A composite tool that creates a template from raw HTML without you wiring up hosting: it validates CM's template rules, exposes the HTML + images on a temporary public URL (a tunnel CLI on PATH — cloudflared, ngrok, or lt — or your own publicUrlBase), calls createTemplate, and tears the tunnel down. No tunnel dependency is bundled; it shells out to whatever you have. Works on local (stdio) deployments only (it reads your local files). It does not send — create and send the campaign separately.

// arguments
{
  "clientId": "...",
  "name": "Spring sale",
  "html": "/abs/path/email.html",          // or the HTML string itself
  "images": [{ "path": "/abs/hero.jpg", "as": "assets/hero.jpg" }],
  "tunnel": "auto"                           // auto | cloudflared | ngrok | lt
}

Embedding the server in your own process

import { createServer } from '@create-send/mcp';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';

const server = createServer({ apiKey: process.env.CREATESEND_API_KEY });
await server.connect(new StdioServerTransport());

Regenerating from the OpenAPI spec

The tool table in src/tools.generated.ts is generated from spec/createsend-openapi.yaml:

npm run generate

To update for a new spec version, replace spec/createsend-openapi.yaml and re-run the script. src/tools.generated.ts is overwritten; the hand-written src/client.ts, src/server.ts, and src/index.ts are not.

Build

npm install
npm run generate
npm run typecheck
npm run build