Google Search Console

What the Google Search Console MCP server does

The Google Search Console MCP server connects an MCP client to the Google Search Console API for read-only investigation. It exposes 20 tools covering accessible properties, Search Analytics, URL Inspection, sitemaps, and derived SEO analyses. No tool submits or deletes sitemaps, changes properties, or requests indexing.

The basic inspection tools provide direct API-oriented data. gsc_list_sites enumerates properties available to the current credentials, while gsc_get_site returns a property's permission level. gsc_query_search_analytics retrieves performance rows, and gsc_inspect_url reports indexing, crawl, mobile usability, AMP, and rich-result information.

The server also includes analysis tools for questions that require comparison or aggregation. These include period-over-period gains and losses, query clustering, cannibalization detection, search-type comparisons, search-appearance trends, indexation watchlists, and large-site URL sampling plans.

How it works

Run the server locally through an MCP client that supports stdio. The process uses Google OAuth credentials and requests the webmasters.readonly scope. A Google account must own or have access to the Search Console properties being queried. gsc_health_check can verify authentication, visible properties, the configured default site, and permission levels without exposing tokens.

Set a default property with GSC_SITE_URL, or pass the property to individual calls. URL-prefix properties use a value such as https://example.com/, including the trailing slash; domain properties use the sc-domain:example.com form. gsc_list_sites helps identify properties available to the account.

URL Inspection calls are limited to 10 URLs. Bulk inspection is sequential and conservative. Search Analytics pagination is bounded by four pages, 100,000 rows, and a 30-second total deadline, with pagination and truncation details reported in the result.

Setup and configuration

The repository release requires Node.js 22.12 or newer. A source-based setup uses the following steps:

git clone --branch v2.0.1 --depth 1 https://github.com/get-mcp-ads/google-search-console-mcp-server.git
cd google-search-console-mcp-server
npm ci
npm run build

Configure the MCP client to launch node with the absolute path to dist/cli.js. In Google Cloud, enable the Google Search Console API and create an OAuth client, using a Desktop app client for local use. Complete the consent flow with an account that can read the target properties and retain the refresh token. A short-lived GSC_ACCESS_TOKEN can be used for a brief trial, but it expires within about an hour.

Claude Desktop, Claude Code, and Cursor are identified as compatible local clients in the project documentation. The hosted option is separate from this repository and is not needed for a local installation.

Tools and capabilities

• List properties and inspect exact access levels.
• Query Search Analytics by supported metrics, dimensions, filters, and search types.
• Compare web, image, video, news, Discover, and Google News performance.
• Check sitemap processing states, warnings, errors, and submitted versus indexed totals.
• Detect data freshness and monitor sitemap and indexation signals.
• Inspect individual URLs or create a bounded inspection plan for larger sites.
• Find performance gains and losses, cluster queries, and identify possible cannibalization.
• Validate metric, dimension, and search-type combinations before querying.

Limitations and notes

Search Analytics responses with dimensions represent the highest-ranked returned rows, not a complete site total. Summing those rows can undercount traffic, especially for large sites and long-tail queries. Query without dimensions when a total is required, and use the same dimensions, row limit, and search type when comparing periods.

Search Console generally retains about 16 months of history, and data can take around three days to settle. Use gsc_get_data_freshness before treating recent figures as final. The server is read-only by design and cannot perform Search Console write operations. Its standalone tools also differ from hosted product features such as MCP Apps or other hosted workflows.