Google Analytics 4 MCP server

Understand what happens after the click.

Explore traffic, conversions, funnels and property configuration through a read-only MCP server.

Watch the demo · What's new · Install · Tool reference · Try hosted getmcpads

Play the 27-second product film

The film demonstrates hosted getmcpads with staged data. Its creative galleries and MCP Apps interface belong to the hosted product. This repository provides the standalone native API tools.

27 read tools · Read-only by design.

Run locally with your own platform credentials and a client that supports stdio MCP, such as Claude Desktop, Claude Code or Cursor. Your requests go directly to the platform. For managed connections, including supported ChatGPT setups, use the hosted option.

What's new

v2.0.0: Native tools and security update · September 20, 2026

• Confirm the current native 27-tool read catalog against the hosted implementation.

• Retain standalone OAuth and redirect protections; no hosted account-selection or application database dependency is introduced.

• Require Node.js 22.12 or newer and check Node 22/24 in CI.

• Update vulnerable dependencies and regenerate the MCP catalog.

Full changelog · Source synchronization details · All releases

Upgrade notes

Requires Node.js 22.12 or newer. CI covers Node 22 and 24. Version 2.0.0 drops Node 18 and 20 support. Hosted creative integrations and MCP Apps UI are outside this release.

Related MCP server: google_analytics

Install this release

This is a GitHub source release. npm and MCP Registry versions are published separately. The commands below select this exact version; unpinned npx examples later in this document select the version currently available on npm.

git clone --branch v2.0.0 --depth 1 https://github.com/getmcpads-com/google-analytics-mcp-server.git
cd google-analytics-mcp-server
npm ci
npm run build

Configure your MCP client to run node with the absolute path to dist/cli.js and the platform credentials documented below.

Prefer a managed connection? Use Google Analytics 4 with hosted getmcpads. Connect your account, select the data your assistant may access and use the hosted MCP connection. See the site for current features and plans.

What you get

Compatibility, checked before the call

GA4 rejects many metric and dimension combinations, and its errors rarely explain which pair is at fault. This server carries the compatibility matrix, so ga4_check_compatibility and ga4_validate_query let the model verify a combination before spending a call and a quota token on it.

Quotas matter here more than on ad platforms: GA4 charges tokens per property per day, and a few careless exploratory queries can exhaust them. ga4_get_property_quotas_snapshot shows what is left.

Read-only, and why it stays that way

There are no write tools, and no environment variable that adds any. Every tool calls a read method of the Data API or the Admin API.

This is not caution for its own sake. A misread report is a wrong answer you can spot. A mistaken write to an analytics property, a deleted audience or an edited data stream, corrupts the record you use to judge everything else, and often silently. The other servers we publish do have write tools, guarded by a mandatory preview. This one has none.

Our ad platform servers with guarded writes: Meta Ads · Google Ads · TikTok Ads

Personal data

Analytics data is not anonymous by default. A GA4 property can carry a userId you set yourself, a device identifier, or a Google Signals pseudonymous ID. An MCP conversation sends whatever a tool returns to a model.

Identifiers are redacted by default wherever this server can return them, and showing them is an explicit opt-in, never the default.

Every one of these accepts includePersonalIdentifiers: true to return raw values, and defaults to false.

Audience export rows are positional, so the redaction reads the export's dimension metadata to find identifier columns. If that metadata is missing, every value in the row is redacted rather than guessing which column is safe. Failing closed is the point.

Report rows are not redacted. ga4_run_report and the other reporting tools return what you asked for. Altering the numbers a report returns would be worse than returning them.

So one decision stays yours: if your property carries a userId you consider personal, do not request it as a report dimension in a conversation whose transcript leaves your machine.

Getting credentials

Three values, obtained once.

1. OAuth client

In a Google Cloud project, enable the Google Analytics Data API and the Google Analytics Admin API, then create an OAuth client under APIs & Services → Credentials. Choose Desktop app for local use. Note the client ID and client secret.

2. Refresh token

Run the OAuth consent flow once, signed in as a Google account with access to your GA4 properties, and keep the refresh token. The analytics.readonly scope is enough, and it is the only one you should grant.

📖 Google OAuth for installed apps

The refresh token does not expire. It is the sensitive value: anyone holding it can mint access tokens indefinitely. Use an OAuth client dedicated to this server so you can revoke it on its own.

3. Property ID, optional

Set GA4_PROPERTY_ID to avoid passing it on every call. Find it in GA4 under Admin → Property Settings, or list them with ga4_list_properties.

Run ga4_health_check as your first call. It verifies the credentials and lists the properties you can actually reach, without printing any secret.

Setup

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "google-analytics": {
      "command": "npx",
      "args": ["-y", "@getmcpads/google-analytics-mcp-server"],
      "env": {
        "GA4_CLIENT_ID": "your-client-id",
        "GA4_CLIENT_SECRET": "your-client-secret",
        "GA4_REFRESH_TOKEN": "your-refresh-token"
      }
    }
  }
}

Restart Claude Desktop. Ask it: "list my Google Analytics properties".

Claude Code

claude mcp add google-analytics --env GA4_CLIENT_ID=... --env GA4_CLIENT_SECRET=... --env GA4_REFRESH_TOKEN=... -- npx -y @getmcpads/google-analytics-mcp-server

Cursor

.cursor/mcp.json in your project, same shape as the Claude Desktop config above.

From source

git clone https://github.com/getmcpads-com/google-analytics-mcp-server.git
cd google-analytics-mcp-server
npm install && npm run build
cp .env.example .env   # then fill in your credentials
npm start

Configuration

Check your setup at any time:

npm run doctor

Tools

Every tool is listed below. See server-card.json for complete parameter and output schemas.

Security

• The client secret and refresh token are never logged, at any log level, or written to disk.

• Three hosts are contacted, and only three: analyticsdata.googleapis.com, analyticsadmin.googleapis.com and oauth2.googleapis.com. A test fails the build if a fourth host appears in the source.

• No fetch follows a redirect. Every outbound call sets redirect: "error", so a redirect cannot forward a bearer token or client secret to another host. A test fails the build if any fetch omits this.

• No telemetry. The server makes no network call other than to Google.

Full policy, including how personal data is handled: SECURITY.md.

Looking for a managed, multi-platform version?

Try hosted GA4 if you want to use this source without operating a local server. getmcpads also connects advertising, Search Console and GA4 through one MCP URL. Source availability and plan limits are listed on the site; connecting an account is still required.

1. Follow the GA4 connection guide.

2. Select the account or property your assistant may read.

3. Connect Claude, ChatGPT or Codex.

4. Try a read-only review: “Compare acquisition channels, landing pages and configured key events. State missing data and do not change anything.”

See the current hosted tool catalogue and pricing before choosing a paid plan. This Apache 2.0 adapter remains independently useful with your own credentials.

Contributing

Issues and pull requests are welcome. See CONTRIBUTING.md. Please read SECURITY.md before reporting anything security-related.

Licence

Apache License 2.0. See also NOTICE.

Google, Google Analytics and GA4 are trademarks of Google LLC. This project is not affiliated with, endorsed by, or sponsored by Google LLC. It is an independent client of a public API.

MCP contracts and desktop bundle

Every tool declares read/write annotations, parameter descriptions and a structured output schema. Successful calls expose the payload as structuredContent.result; errors retain isError: true. The generated server card contains definitions only.

Run npm run bundle -- /path/to/output to build a .mcpb desktop bundle from the current catalog. Credentials are entered locally during installation. This server remains read-only.

More from getmcpads

Meta Ads · Google Ads · Google Search Console · TikTok Ads · Pinterest Ads · X Ads

Maintained by Emmanuel at getmcpads. Questions: hello@getmcpads.com.