Wascer GTM MCP Server

A remote MCP (Model Context Protocol) server for Google Tag Manager, powered by Wascer. Enables AI assistants like Claude, Cursor, and others to manage GTM accounts, containers, tags, triggers, variables, and more.

Features

• Google OAuth authentication with full GTM API scopes

• Service Account support (optional) via gtm_setup tool for platform-level access

• 11 tools covering the complete GTM workflow: accounts, containers, workspaces, tags, triggers, variables, versions, built-in variables, clients, and more

• Remote-first architecture on Cloudflare Workers with Durable Objects

Related MCP server: unboundai-gtm-mcp-server

Available Tools

Authentication Modes

Mode 1: Google OAuth (default)

Users authenticate with their Google account. The MCP server requests GTM permissions during login. Users can access any GTM account their Google account has permissions on.

Mode 2: Service Account (optional)

For platform-level access, users can call gtm_setup with a Google Service Account JSON. This overrides the OAuth token for the session and grants access to all GTM accounts the Service Account has been granted permissions on.

Setup

Claude Desktop

Open Settings -> Developer -> Edit Config and add:

{
  "mcpServers": {
    "wascer-gtm": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://your-worker-url.com/mcp"
      ]
    }
  }
}

Claude Code (CLI)

claude mcp add wascer-gtm -- npx mcp-remote https://your-worker-url.com/mcp

Cursor / VS Code

Add to your MCP settings:

{
  "mcpServers": {
    "wascer-gtm": {
      "command": "npx",
      "args": ["mcp-remote", "https://your-worker-url.com/mcp"]
    }
  }
}

After connecting, a browser window will open for Google OAuth. Complete the login to grant GTM access.

Local Development

Prerequisites

• Node.js 18+

• A Google Cloud OAuth app (GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET)

• Add http://localhost:3333/callback as an authorized redirect URI in the Google Cloud Console

Setup

git clone https://github.com/wascer/wascer-gtm-mcp-server.git
cd wascer-gtm-mcp-server
npm install

Create a .dev.vars file:

GOOGLE_CLIENT_ID=your-client-id
GOOGLE_CLIENT_SECRET=your-client-secret
COOKIE_ENCRYPTION_KEY=any-secret-string
WORKER_HOST=http://localhost:3333

Run

npx wrangler dev

The server starts at http://localhost:3333.

Test with MCP Inspector

npx @modelcontextprotocol/inspector

Connect using STDIO transport with command npx and arguments mcp-remote http://localhost:3333/mcp.

Run Tests

npm test

Build

npm run build

Deployment

Cloudflare Workers

1. Create a KV namespace:

wrangler kv namespace create OAUTH_KV

2. Update wrangler.jsonc with the KV namespace ID

3. Set secrets:

wrangler secret put GOOGLE_CLIENT_ID
wrangler secret put GOOGLE_CLIENT_SECRET
wrangler secret put COOKIE_ENCRYPTION_KEY
wrangler secret put WORKER_HOST

4. Deploy:

npm run deploy

Troubleshooting

MCP Server Name Length Limit

Some MCP clients have a 60-character limit for the combined server name + tool name. Use a short server name like wascer-gtm.

Clearing Auth Cache

mcp-remote stores credentials in ~/.mcp-auth. To reset:

rm -rf ~/.mcp-auth

Then restart your MCP client.

Google OAuth Consent Screen

If you see "Access blocked" during login, make sure your Google Cloud OAuth app has the GTM API scopes enabled and the app is configured for the correct user type (internal or external with test users).

License

Apache-2.0