ga-mcp-full

A Model Context Protocol server that exposes the full Google Analytics 4 surface — both the Admin API (property & configuration management) and the Data API (reporting) — to MCP-compatible clients such as Claude Code, Claude Desktop, Cursor, Zed, and anything else that speaks stdio MCP.

Thirty-plus tools cover:

• Property management — create, list, update, delete, archive, retention settings
• Custom dimensions & metrics — create, list, update, archive
• Audiences — create, list, archive
• Key events — CRUD
• Data streams — CRUD
• Linked accounts — Firebase, BigQuery, Google Ads
• Measurement Protocol secrets — CRUD
• Reporting — standard reports, realtime reports, metadata discovery

Install

The preferred install is via the Robworks Claude Code Plugins marketplace:

pip install ga-mcp-full
claude plugin marketplace add ringo380/robworks-claude-code-plugins
claude plugin install ga-mcp-full@robworks-claude-code-plugins

Why a two-step install: the plugin just wraps the Python CLI — it does not bundle the Python interpreter. You install the CLI once (pip), then the plugin registers the MCP server, slash commands, and SessionStart hook with Claude Code.

Alternative: direct from the plugin repo

If you prefer to install the plugin directly from this repo without going through the Robworks marketplace:

pip install -e /path/to/ga-mcp-full
claude plugin marketplace add ringo380/ga-mcp-full   # uses the plugin repo as its own marketplace (once marketplace.json is added here)
claude plugin install ga-mcp-full

Note: this repo currently only ships a plugin.json, not a marketplace.json. Use the Robworks marketplace above as the install path.

Alternative: standalone MCP (no Claude Code plugin)

pip install -e /path/to/ga-mcp-full

Then add to your MCP client config. Example for Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "ga-mcp-full": {
      "command": "ga-mcp-full",
      "args": ["serve"]
    }
  }
}

For Cursor, Zed, or other clients, consult their MCP integration docs — the server command and arg pair is always ga-mcp-full serve.

Requirements

• Python 3.10 or newer
• Google Analytics 4 property you have Editor or Administrator access to (the OAuth flow requests analytics.edit for Admin API writes and analytics.readonly for Data API reporting)

That's it — no Google Cloud Console visit needed for the default install. A public Desktop OAuth client is bundled with the package.

Authentication

The server resolves credentials in this order (see ga_mcp/auth.py):

1. Cached OAuth tokens at ~/.config/ga-mcp/credentials.json. Refresh tokens renew expired access tokens silently.
2. Application Default Credentials — but only if the ADC token was actually granted the analytics.edit scope. A plain gcloud auth application-default login does not include this scope; you must run gcloud auth application-default login --scopes=https://www.googleapis.com/auth/analytics.edit. (Append ,https://www.googleapis.com/auth/analytics.readonly if you also want the reporting tools, which call the GA4 Data API and require the read-only scope.) If the scope is missing, resolution falls through with a clear error.

If neither source yields a working credential, the server surfaces a single actionable message telling you to run /ga-mcp-full:auth-login. The MCP server itself never opens a browser — the OAuth flow is driven by the CLI command (ga-mcp-full auth login) or its slash-command wrapper, since the MCP stdio subprocess has no attached terminal.

First-time setup (majority path)

1. Install: pip install ga-mcp-full (or pip install -e . for a dev checkout).
2. Run /ga-mcp-full:auth-login in Claude Code (or ga-mcp-full auth login at the shell).
3. Approve the requested scopes (analytics.edit + analytics.readonly) when the browser opens.
4. The refresh token lands in ~/.config/ga-mcp/credentials.json (mode 0600).

The OAuth flow uses PKCE (S256) and binds the loopback redirect to a random free port on 127.0.0.1, per Google's installed-app OAuth guidance. No prior Google Cloud Console client configuration is required.

You'll see an "unverified app" warning — this is expected for now

The bundled OAuth client is a real production app owned by this project, but Google's brand verification for sensitive scopes (like analytics.edit) takes several business days the first time. Until verification clears, the consent screen displays:

Google hasn't verified this app The app is requesting access to sensitive info in your Google Account…

To proceed: click Advanced → Go to ga-mcp-full (unsafe). The warning is Google's default posture for unverified apps — it doesn't indicate anything is actually wrong with the tool. Review the privacy policy if you want to know exactly what data is accessed (it's all local, per-user, per the scope you approved). The warning will disappear once Google's review completes.

Users who want to avoid the warning entirely can use their own OAuth client — see Power users below.

Power users: use your own OAuth client

If you want quota isolation, a custom consent screen, or to avoid the verification-window warning:

1. Create an OAuth 2.0 Client ID of type Desktop app at Google Cloud Console → APIs & Services → Credentials in a project where the Google Analytics Admin API and Data API are enabled.
2. Override the bundled defaults via either:
   • Env vars in your shell profile: GA_MCP_CLIENT_ID=... and GA_MCP_CLIENT_SECRET=...
   • Or drop the downloaded JSON at ~/.config/ga-mcp/client_secrets.json (mode 0600).
3. Run ga-mcp-full auth login as usual — the env vars / legacy JSON take precedence over the bundled client.

Already using gcloud?

If you have gcloud auth application-default login --scopes=https://www.googleapis.com/auth/analytics.edit,https://www.googleapis.com/auth/analytics.readonly configured, the server auto-detects ADC and skips OAuth entirely. No further setup needed. (The analytics.readonly scope is only needed for the reporting tools; analytics.edit alone is enough for the Admin API.)

Why not the /mcp Authenticate button?

Claude Code's /mcp menu OAuth flow only applies to HTTP-transport MCP servers per the MCP 2025-06-18 authorization spec (stdio servers "SHOULD NOT" use it). ga-mcp-full is stdio, so auth is handled out-of-band via the bundled slash commands and the SessionStart hook. An HTTP transport mode that participates in the /mcp OAuth flow is a possible future addition.

Bundled Claude Code components

Slash commands

SessionStart hook

hooks/scripts/ensure-auth.sh runs at the start of every Claude Code session (startup|resume|clear|compact). It is non-blocking and only writes to stderr — if the CLI is installed but unauthenticated, it prints a one-line hint pointing at /ga-mcp-full:auth-login; if the CLI is not on PATH at all, it points at pip install ga-mcp-full. It never exits non-zero and never blocks the session.

MCP tools

After plugin install the server registers as plugin:ga-mcp-full:ga-mcp-full and tools surface as mcp__ga-mcp-full__*. Browse them with /mcp inside Claude Code.

CLI reference

ga-mcp-full                 # Start the MCP server over stdio
ga-mcp-full serve           # Same as above (explicit)
ga-mcp-full auth login      # OAuth browser flow; caches refresh token
ga-mcp-full auth logout     # Clear cached credentials
ga-mcp-full auth status     # Show current auth state and credential source
ga-mcp-full --help          # Full help

Development

git clone https://github.com/ringo380/ga-mcp-full.git
cd ga-mcp-full
python -m venv .venv && source .venv/bin/activate
pip install -e .
ga-mcp-full auth status

Project layout:

ga_mcp/
├── server.py          # MCP server entrypoint
├── cli.py             # CLI dispatcher
├── auth.py            # OAuth + credential resolution
└── tools/
    ├── admin/         # 11 modules — property, custom dimensions/metrics, audiences, etc.
    ├── reporting/     # 3 modules — core, realtime, metadata
    └── utils.py

Public Claude Code artifacts: .claude-plugin/plugin.json, .mcp.json, commands/, hooks/.

Issues and PRs welcome at https://github.com/ringo380/ga-mcp-full/issues.

Troubleshooting

Related

• Robworks Claude Code Plugins marketplace — catalogs this plugin alongside future plugins from @ringo380
• Claude Code plugin docs
• MCP specification
• Google Analytics Admin API
• Google Analytics Data API

License

MIT — see LICENSE.