How to Connect Claude to Google Search Console With a Local MCP Server

A local Model Context Protocol server can give Claude a structured way to query Search Console without repeatedly exporting CSV files. The open-source mcp-gsc project exposes tools for performance analysis, comparisons, property access, URL inspection, indexing checks and sitemaps. Claude decides which tool fits the question, the local server calls the Google API and the result returns to the conversation.

• A Google account with access to the required Search Console properties.
• A Google Cloud project with the Search Console API enabled.
• Claude Desktop or another MCP-compatible client.
• The uv and uvx runtime recommended by the project, or a local Python environment for the advanced clone method.
• Permission to install and connect local software on the device.

In Google Cloud Console, create or select a project and enable the Search Console API. For individual use, the repository recommends an OAuth desktop client. Configure the consent screen, create a Desktop app OAuth client and download the JSON file to a permanent private location.

For team automation, the server also supports a service account. Create the account and JSON key, then add the service account email to the required Search Console property with the minimum appropriate permission. The user who runs queries still needs access to the data being requested.

The project's current recommended installation uses uvx, which downloads and runs the published mcp-search-console package. Install uv using the method documented for your operating system, confirm that uvx is available and note its full executable path. Desktop applications may not inherit the same PATH as your terminal, so the full path avoids common spawn errors.

Open the Claude Desktop MCP configuration file and add a server entry named gscServer. Set command to the full uvx path, args to mcp-search-console and add either GSC_OAUTH_CLIENT_SECRETS_FILE or the service-account variables described in the README. Every file path should be absolute. Do not paste the credential JSON contents into the configuration or a prompt.

For OAuth, the first tool call opens a browser sign-in and consent flow. After approval, the token is cached locally. Ask Claude to list available Search Console properties. If that fails, call the server's get_capabilities tool to inspect authentication status and available tools.

• Compare clicks and impressions for the last 28 days with the previous 28 days by page.
• Find non-branded queries in positions 4 to 15 with high impressions and below-average CTR.
• Show which pages lost clicks and identify the query groups responsible.
• Inspect whether a priority URL is indexed and report the returned status.
• List submitted sitemaps and flag errors without changing them.

Search Console data has limits, privacy thresholds and reporting delays. Ask Claude to state the date range, dimensions, filters, row limits and comparison method. Check important findings in Search Console or the GSC Dashboard before acting.

OAuth is usually easier for a single analyst and follows that user's property permissions. A service account is better suited to controlled automation or shared infrastructure, but it creates another identity that must be managed and explicitly added to properties. Rotate credentials according to organisational policy and remove access when the workflow is retired.

The project disables add-site, delete-site and delete-sitemap operations by default. Leave that safeguard in place for analysis. Run the server locally over standard input and output unless you deliberately need a remote transport and understand its authentication and network exposure.