MCP server, Google Analytics 4 + Search Console

A remote Model Context Protocol (MCP) server, deployed on Cloudflare Workers, that exposes read-only Google Analytics 4 and Google Search Console data to MCP clients such as Claude Desktop, Cowork, Lovable Agent, and the MCP Inspector.

Built for the Rablab agency to power dashboards and content reporting workflows. Open for forks.

Features

• 17 read-only tools across Google Analytics 4 and Search Console

• OAuth 2.0 user authentication via Google with analytics.readonly and webmasters.readonly scopes

• Automatic refresh token, no need to reconnect every hour

• Works with Dynamic Client Registration (DCR) for compatibility with MCP Inspector, Cowork, Lovable, etc.

• Email and domain allowlist via ALLOWED_EMAILS and ALLOWED_DOMAINS secrets, so only authorized accounts can mint an MCP token

• Branded access-denied page for users outside the allowlist (Rablab colors)

• Multi-user friendly, each authorized user authenticates with their own Google account

• Deployed once on Cloudflare Workers, available to a whole team

Related MCP server: SEO Analytics MCP Server

Available tools

Google Analytics 4 (12 tools, all read-only):

• ga4_list_account_summaries, lists all accounts and properties accessible to the user

• ga4_list_properties, lists properties under a specific account

• ga4_get_property_details, returns timezone, currency, industry, etc.

• ga4_list_data_streams, lists web, iOS, Android streams

• ga4_get_metadata, lists all dimensions and metrics available on a property

• ga4_check_compatibility, validates dimension and metric combos

• ga4_run_report, runs a custom report (sessions, conversions, etc. by dimension and date)

• ga4_run_realtime_report, runs a realtime report (last 30 minutes)

• ga4_run_pivot_report, pivot tables

• ga4_batch_run_reports, up to 5 reports in one API call

• ga4_list_key_events, lists key events and conversions

• ga4_list_conversion_events, legacy conversion events

• ga4_list_custom_dimensions, ga4_list_custom_metrics

• ga4_list_audiences, ga4_list_google_ads_links, ga4_list_firebase_links

Google Search Console (5 tools, all read-only):

• gsc_list_sites, lists all GSC properties accessible to the user

• gsc_query_search_analytics, clicks, impressions, CTR, position by dimensions and date

• gsc_list_sitemaps, lists sitemaps submitted

• gsc_get_sitemap, details of a specific sitemap

• gsc_inspect_url, URL inspection (indexation status, last crawl, mobile usability)

No write actions are exposed. Worst case if credentials leak, the attacker can only read GA4 and GSC data they already have access to.

Architecture

MCP client (Claude, Cowork, Lovable, Inspector)
│
│ MCP over SSE
▼
Cloudflare Worker (this repo)
├── @cloudflare/workers-oauth-provider
│    handles MCP OAuth + Dynamic Client Registration
├── google-handler.ts
│    handles Google OAuth flow + email allowlist + refresh token
└── 17 MCP tools
│
▼
Google APIs (analyticsdata, analyticsadmin, searchconsole, webmasters)

Setup for your own deployment

If you fork this repo to deploy your own MCP server.

1. Google Cloud setup

In your Google Cloud project:

1. Enable the following APIs:

• Google Analytics Admin API

• Google Analytics Data API

• Google Search Console API

2. Configure the OAuth Consent Screen (Audience: External, mode Test, add yourself as a test user).

3. Create an OAuth 2.0 Client ID, type Web Application:

• Authorized JavaScript origin: https://<your-worker-name>.<your-subdomain>.workers.dev

• Authorized redirect URI: https://<your-worker-name>.<your-subdomain>.workers.dev/callback

4. Note the Client ID. Click Add secret, copy the Client Secret value (shown only once).

2. Cloudflare setup

1. Create a KV namespace, name it OAUTH_KV. Note its ID.

2. Edit wrangler.jsonc:

• name: your worker name (this controls the URL)

• kv_namespaces[0].id: replace with your KV namespace ID

3. Deploy

npm install --legacy-peer-deps

# Set the secrets (you'll be prompted to paste each value)
npx wrangler secret put GOOGLE_CLIENT_ID
npx wrangler secret put GOOGLE_CLIENT_SECRET
npx wrangler secret put COOKIE_ENCRYPTION_KEY # any random 32-char hex string, e.g. openssl rand -hex 32
npx wrangler secret put HOSTED_DOMAIN # press Enter for empty (accept any Google Workspace domain)
npx wrangler secret put ALLOWED_EMAILS # comma-separated emails, e.g. alice@example.com,bob@example.com
npx wrangler secret put ALLOWED_DOMAINS # optional, comma-separated domains, e.g. example.com

# Deploy
npx wrangler deploy

At least one of ALLOWED_EMAILS or ALLOWED_DOMAINS must be set, otherwise no one can sign in (see Access control below).

4. Access control: ALLOWED_EMAILS and ALLOWED_DOMAINS

After a user signs in with Google, the worker checks their email against two Cloudflare secrets before issuing an MCP token:

• ALLOWED_EMAILS: comma-separated list of full email addresses allowed to authenticate (case-insensitive). Example: alice@example.com,bob@example.com.

• ALLOWED_DOMAINS: comma-separated list of domains. Any email ending with @domain is allowed (case-insensitive). Example: example.com,partner.com.

The two lists are additive: an email is allowed if it matches either one.

If neither secret is set, the worker rejects every sign-in. This is intentional. Without an allowlist, any Google account could complete the OAuth flow and mint a token, burning the deployment.

Unauthorized users get a branded "Acces refuse" 403 page (Rablab colors) instead of a token. No error leaks to the MCP client.

To update the allowlist after deployment:

# Replace the value (you'll be prompted for the new full list)
npx wrangler secret put ALLOWED_EMAILS

The change takes effect on the next sign-in. Existing MCP tokens stay valid until they expire; to revoke immediately, purge the OAUTH_KV namespace.

5. Test with MCP Inspector

npx @modelcontextprotocol/inspector@latest

Open http://localhost:6274. Set Transport Type to SSE, URL to https://<your-worker>.workers.dev/sse, click Connect. A Google OAuth flow opens, approve the scopes. Then try gsc_list_sites to confirm it works.

If you sign in with an unlisted email, you should see the "Acces refuse" 403 page instead. Use this to validate the allowlist before sharing the worker URL with your team.

6. Connect to Claude Desktop

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

{
"mcpServers": {
"ga4-gsc": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://<your-worker>.workers.dev/sse"]
}
}
}

Restart Claude Desktop. The first time you call a tool, an OAuth flow opens in your browser.

Multiple Google accounts on the same machine. Claude Desktop deduplicates MCP servers by URL, so you cannot add the same /sse endpoint twice. The worker supports multi-account via the ALIAS_PATHS env variable: each alias is a separate URL that Claude treats as a distinct connector (e.g. /sse-secondary). By default ALIAS_PATHS is unset so the worker behaves like a normal single-account MCP.

Path-based access control (optional). You can also restrict each path to specific authorized emails via the PATH_EMAIL_MAP env variable, a JSON map of path -> [authorized emails]. Useful when you want an alias path usable by one specific user only (e.g. a test path isolated from team usage, or strict separation between account holders in a transition setup). If PATH_EMAIL_MAP is unset, no path-level restriction applies and any email in ALLOWED_EMAILS can use any path.

See MULTI-ACCOUNT.md for step-by-step scenarios (add a route, remove a route, fork the repo, restrict paths to specific users, troubleshoot).

7. Connect at organization level (recommended)

If you have an Anthropic organization (Claude Team or Enterprise), add the worker URL as a custom connector at the organization level. Every team member gets access in Cowork, Claude Desktop, and claude.ai web without local config. The allowlist enforces who can actually authenticate.

Local development

npm install --legacy-peer-deps
cp .dev.vars.example .dev.vars
# Fill in the values in .dev.vars (do not commit this file)
npx wrangler dev

For local dev, set ALLOWED_EMAILS (and optionally ALLOWED_DOMAINS) in .dev.vars too, otherwise the local worker rejects all sign-ins.

CI/CD with Cloudflare Workers Builds

This repo is set up to auto-deploy on push to main via Cloudflare Workers Builds. Connect this GitHub repository in your Cloudflare Workers dashboard:

• Build command: npm install --legacy-peer-deps

• Deploy command: npx wrangler deploy

• Production branch: main

Each commit on main rebuilds and redeploys the worker. Other branches build preview workers at distinct URLs.

Secrets (GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, COOKIE_ENCRYPTION_KEY, HOSTED_DOMAIN, ALLOWED_EMAILS, ALLOWED_DOMAINS) are set once with wrangler secret put and persist across deploys.

Security notes

• All Google API calls are read-only. No write scope is requested.

• Tokens (access and refresh) are stored encrypted in the OAuth provider's KV store and as Durable Object props, never logged.

• The .dev.vars file (containing local secrets for development) is gitignored.

• Access is restricted by the ALLOWED_EMAILS and ALLOWED_DOMAINS allowlist, checked server-side after Google OAuth. Without at least one of these secrets set, the worker rejects all sign-ins by design.

• Each authorized user authenticates with their own Google account, so the worker only has access to the GA4 and GSC properties that user already has access to. No service account, no shared credentials.

• Set the OAuth Consent Screen to Test mode and only add trusted test users until you complete Google verification (only required for >100 users).

Credits

This project is forked from and inspired by bighadj22/cloudflare-mcp-google-oauth-analytics, which provided the original Cloudflare Workers + Google OAuth + MCP boilerplate. The Rablab fork:

• Replaces the dummy add tool with 17 production-ready Google Analytics 4 and Search Console read-only tools

• Migrates from the deprecated mcp-analytics package to the native agents/mcp SDK from Cloudflare

• Adds access_type=offline + automatic refresh token handling so users don't reconnect every hour

• Fixes Google OAuth token exchange request body to use snake_case as required by Google

• Adds an email/domain allowlist (ALLOWED_EMAILS and ALLOWED_DOMAINS) enforced server-side after Google OAuth, with a branded access-denied page for unauthorized accounts

License

MIT, see LICENSE.