GA4 MCP Server
# GA4 MCP Server
[](https://modelcontextprotocol.io)
[](https://python.org)
[](LICENSE)
A Model Context Protocol (MCP) server that connects Claude Desktop (and other MCP clients) to the Google Analytics 4 Data API — run reports, view top pages, analyze traffic sources, check realtime users, and compare periods.
## Features
### 7 Analytics Tools
| Tool | Description |
|------|-------------|
| `ga4_list_properties` | List all GA4 properties accessible by the service account |
| `ga4_get_report` | Run a custom report with any dimensions and metrics |
| `ga4_get_top_pages` | Top pages by sessions with bounce rate and duration |
| `ga4_get_traffic_sources` | Traffic breakdown by source/medium |
| `ga4_get_conversions` | Conversion events and their counts |
| `ga4_get_realtime` | Realtime active users and top pages |
| `ga4_compare_periods` | Compare current vs previous period with percentage changes |
### Built-in Reliability
- **Token-bucket rate limiter** — respects GA4's 10 concurrent request limit
- **Auto-retry on 429/5xx** — exponential backoff (2s, 4s, 8s) up to 3 retries
- **Lazy imports** — defers heavy Google client libraries for fast MCP handshake
- **Actionable error messages** — guides users to fix permission, auth, and input errors
## Quick Start
### Prerequisites
- Python 3.11+
- A Google Cloud service account with GA4 access
- Claude Desktop (or any MCP-compatible client)
### Set Up Google Analytics Access
1. Create a [Google Cloud service account](https://console.cloud.google.com/iam-admin/serviceaccounts)
2. Enable the **Google Analytics Data API** and **Google Analytics Admin API** in your project
3. Download the service account JSON key file
4. In GA4: go to **Admin > Property Access Management** and add the service account email as a **Viewer**
### Installation
Run the published package without a global install:
```bash
uvx --from luminarylane-ga4-mcp luminarylane-ga4-mcp
```
The package is published on [PyPI](https://pypi.org/project/luminarylane-ga4-mcp/). Alternatively, install it with `pip install luminarylane-ga4-mcp`.
### Configuration
**Claude Desktop (`claude_desktop_config.json`):**
```json
{
"mcpServers": {
"ga4": {
"command": "uvx",
"args": ["--from", "luminarylane-ga4-mcp", "luminarylane-ga4-mcp"],
"env": {
"GA4_CREDENTIALS_PATH": "/path/to/service_account_credentials.json",
"GA4_PROPERTY_ID": "123456789"
}
}
}
}
```
**Environment variables:**
| Variable | Required | Description |
|----------|----------|-------------|
| `GA4_CREDENTIALS_PATH` | Yes | Path to service account JSON key file |
| `GA4_PROPERTY_ID` | No | Default property ID (can also pass per-tool) |
For local development only, credential files may be stored in paths excluded by `.gitignore`; never commit them. Prefer the explicit environment-variable path above.
## Usage Examples
Once configured, ask Claude to:
- "What are my top pages this month?"
- "Show me traffic sources for the last 7 days"
- "How does this week compare to last week?"
- "Are there any active users on the site right now?"
- "Run a report on sessions by country for the last 90 days"
- "What conversion events fired this week?"
- "List all GA4 properties I have access to"
## Rate Limits
GA4 Data API allows 10 concurrent requests per property. The server handles this with a client-side token bucket and exponential backoff retries.
## Troubleshooting
### Permission denied (403)
The service account doesn't have access to the GA4 property. Go to **GA4 Admin > Property Access Management** and add the service account email as a Viewer.
### Property not found (404)
Property IDs are numeric (e.g., `123456789`), not the measurement ID (`G-XXXXXXX`). Use `ga4_list_properties` to find valid IDs.
### Authentication failed (401)
Check that `GA4_CREDENTIALS_PATH` points to a valid service account JSON key file.
### Invalid dimension/metric (400)
Check the [GA4 Dimensions & Metrics Explorer](https://developers.google.com/analytics/devguides/reporting/data/v1/api-schema) for valid names. Common mistakes: `sessions` not `session`, `sessionSource` not `source`, `pagePath` not `page`.
## Contributing
See [CONTRIBUTING.md](CONTRIBUTING.md). Never commit service-account credentials or OAuth tokens.
## Security
Report vulnerabilities privately using [GitHub's private vulnerability reporting](SECURITY.md); do not include credentials in public issues.
## License
MIT License — see [LICENSE](LICENSE) for details.
## Acknowledgments
- [Anthropic](https://anthropic.com) for the MCP specification
- [Google Analytics Data API](https://developers.google.com/analytics/devguides/reporting/data/v1) for the reporting API
TDQS
Scored across 7 tools
Each special-purpose tool targets a distinct analytical view (top pages, traffic sources, conversions, realtime, period comparison), and ga4_get_report is explicitly generic. While ga4_get_report could theoretically replicate the specialized tools, its purpose is clearly different, so confusion is unlikely.
All tools follow the consistent ga4_<verb>_<noun> pattern with lowercase and underscores. Verbs are list, get, and compare, which is a predictable and uniform convention across the set.
With 7 tools, the server is well-scoped for a GA4 analytics wrapper. It covers property discovery and the most common report types without unnecessary bloat or excessive minimalism.
The tool set covers property listing, common report dimensions, custom queries, realtime data, and period-over-period comparison. Advanced GA4 features like funnel analysis or user exploration are absent, but these are beyond the typical expectation for an analytics MCP server.