mcp-google-gsc

MCP server for Google Search Console -- search analytics, URL inspection, and site management via Claude.

Features

• Search Analytics -- Query clicks, impressions, CTR, and position with flexible dimension filters (query, page, device, country, date)
• URL Inspection -- Check indexing status, mobile usability, and rich results for any URL
• Site Listing -- List all verified Search Console properties accessible to your authenticated account (OAuth user or service account)
• Multi-Client Support -- Manage multiple GSC properties with per-directory config mapping

Installation

From npm

npm install mcp-google-gsc

From source

git clone https://github.com/mharnett/mcp-search-console.git
cd mcp-google-gsc
npm install
npm run build

Configuration

Security: Never share your .mcp.json file or commit it to git -- it may contain API credentials. Add .mcp.json to your .gitignore.

mcp-gsc supports two authentication modes. Pick whichever fits your setup. Neither requires any file living at a hardcoded machine-local path -- credentials come from environment variables (or a config.json you create).

Which mode to use:

• Service Account (Mode B) -- recommended for unattended / server / headless use. A service account has no interactive login to expire or re-consent, so it is the right fit for always-on deployments. The one setup requirement is that the service account's email must be granted access on each Search Console property you want to query (see Mode B).
• User OAuth (Mode A) -- for personal / interactive use, where you want to authorize with your own Google login.

Precedence when both are configured: if a service-account keyfile is explicitly set (GOOGLE_APPLICATION_CREDENTIALS, or credentials_file in config.json), it wins over any OAuth refresh token or stored OAuth credentials. If neither is configured, the server fails loudly at startup with an onboarding message rather than silently guessing -- there is no machine-local default and no silent runtime failover between modes.

Mode A: User OAuth (bring your own Google account)

Use this if you want to authorize with your own Google login (the account that has Search Console access). Best for personal / interactive use.

1. In the Google Cloud Console, create an OAuth 2.0 Client ID of type Desktop app and enable the Search Console API. (For a Desktop-app client, Google accepts any http://localhost loopback redirect -- you do not need to pre-register a port.)
2. Export your client credentials:

   export GOOGLE_GSC_CLIENT_ID=...apps.googleusercontent.com
   export GOOGLE_GSC_CLIENT_SECRET=...
   

3. Mint a refresh token (opens your browser, uses PKCE + access_type=offline):

   node get-refresh-token.cjs
   

   Do not redirect this command's stdout to a shared log -- it prints the refresh token to stdout by design.

4. Copy the printed line into your environment:

   export GOOGLE_GSC_REFRESH_TOKEN=...
   

The server reads GOOGLE_GSC_CLIENT_ID, GOOGLE_GSC_CLIENT_SECRET, and GOOGLE_GSC_REFRESH_TOKEN from the environment at runtime.

Alternatively, run the guided helper npx mcp-gsc-auth, which performs the same PKCE OAuth flow, lets you pick a default Search Console property, and writes the result to a per-user credentials file.

Mode B: Service Account (recommended for unattended / server use)

Use this for server / headless / always-on contexts -- it is the recommended path when no human is present to complete or refresh an interactive login.

1. Create a Google Cloud service account with Search Console API access and download its JSON key file.
2. Grant the service account's email access on each Search Console property you want to query (add it as a user in Search Console). Without this grant on the property, the service account can authenticate but will see no sites.
3. Point the server at the key file via an environment variable (no hardcoded path):

   export GOOGLE_APPLICATION_CREDENTIALS=/path/to/your/service-account-key.json
   

OAuth scope

Both modes request a single, read-only scope: https://www.googleapis.com/auth/webmasters.readonly. All four tools are reads -- nothing writes -- so the server never asks for read/write access.

The scope is defined once in config.json under oauth.scope (see config.example.json). The OAuth helper and the runtime read the same value, so they never drift. If config.json is absent (e.g. a fresh install), the committed read-only default is used.

Multi-client config (optional)

To map working directories to Search Console properties, create a config.json in the project root (see config.example.json):

{
  "oauth": {
    "scope": "https://www.googleapis.com/auth/webmasters.readonly"
  },
  "clients": {
    "my-project": {
      "name": "My Project",
      "folder": "/path/to/project",
      "site_url": "https://example.com/"
    }
  }
}

Usage

Add to your Claude Code .mcp.json:

{
  "mcpServers": {
    "gsc": {
      "command": "node",
      "args": ["/path/to/mcp-gsc/dist/index.js"]
    }
  }
}

Or if installed globally:

{
  "mcpServers": {
    "gsc": {
      "command": "npx",
      "args": ["mcp-google-gsc"]
    }
  }
}

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

Tools

gsc_search_analytics

Supports dimensions: query, page, device, country, date. Filter by any dimension with operators like equals, contains, notContains. Date range defaults to the last 28 days.

gsc_inspection

Returns index coverage, crawl status, mobile usability verdict, and rich result details for a specific URL within a property.

Architecture

• Resilience -- Uses cockatiel for retry with exponential backoff and circuit breaker patterns on all Google API calls
• Logging -- Structured logging via pino with configurable log levels
• Response Handling -- Responses truncated at 200KB to stay within MCP transport limits

License

MIT -- see LICENSE.

Built by Mark Harnett / drak-marketing