Skip to content

Search Console MCP documentation

Use Search Console MCP with Claude, Codex or another AI assistant to understand your latest SEO changes and get a prioritized list of fixes. Ask in plain language, without needing SEO expertise: “Analyze my site’s recent performance and tell me what to fix first.”

The server retrieves search metrics and audits pages; your assistant explains traffic changes, finds ranking opportunities and suggests changes to titles, content, internal links or technical SEO. Connect your accounts once, review the suggestions, then rerun the analysis to measure their effect. Recurring checks require a client automation or scheduler, and ranking improvements are not guaranteed.

Your problemOpen the routeWhat you get
I’m new to SEO: where do I start?Assess my site and learn what mattersA starting assessment, three priorities and what to measure
My traffic is droppingAnalyze recent changesA diagnosis and the affected pages, with supporting metrics
I want better search rankingsFind ranking opportunitiesA prioritized list of suggested fixes
My pages are hard to findCheck indexing and technical SEODetected issues and suggested corrections

Each route includes a prompt to copy into Claude or Codex, the data needed and an illustrative result. The copied text asks the assistant to use Search Console MCP and verify its connection before analysis. It includes links to the installation guide, Google setup and the matching GitHub example so the assistant can guide any missing setup. Replace the example site before sending the prompt.

If you are new to SEO, choose the first route. Your assistant explains what to analyze, how to interpret the findings and which metrics to follow. You can start with public pages; connect Google Search Console for search performance data. GA4 and Bing are optional.

Three separate evidence streams converge on a bounded analysis workspace.
Google, Bing, and public-page evidence stay distinct while the assistant analyzes them in one bounded workspace.
  1. Install the server without launching duplicate long-running processes.
  2. Connect Google and verify the properties exposed to the configured identity.
  3. Connect Bing and keep the account API key separate from IndexNow host keys.
  4. Run a starter prompt or choose an example workflow.
  5. Interpret the evidence before submitting a write action.

The server can return provider responses, fetched public-page data, and calculations tied to explicit windows. An accepted submission proves that a provider accepted a request. It does not prove a later crawl or indexation.

GoogleSearch performance, inspection, GA4 and CrUX when configured.
BingWebmaster performance, crawl, feeds, backlinks and bounded submissions.
Public pagesHTML, robots, sitemaps, metadata, schema and internal links.

Read the architecture, review the changelog, or inspect the source repository.

For explicit Google comparison windows and bounded destination HTTP checks, use bounded audit workflows. These tools were added in release 1.3.0. Version 1.3.1 retains the 85-tool registry and adds startup family selection, CLI JSON string lists and SEO/Bing feedback fixes.

Version 1.6.0 retains the 96 tools introduced in 1.5.0, including bounded local logs, stored crawl observations, inventory differences and an indexing evidence matrix. Main-content extraction is optional and richer traffic references remain descriptive. Read editorial workflows for the draft and mechanical comparison boundaries.

Source workflows and independent evaluation

Section titled “Source workflows and independent evaluation”

The bounded native audit runs an explicit acquisition plan with nine selectable source-only roles, durable attempt budgets and per-invocation output limits. The experimental Python APIs are packaged in 1.6.0; the runner scripts and workspace skills require a repository checkout and are not installed console commands. Offline host runs do not establish authenticated provider acquisition or expert report quality.

Use the human evaluation intake to collect authorized independent labels and freeze targets. Query and pair evaluation and local extraction comparison assess separate inputs and metrics. The optional extraction profile has been available since 1.5.0; its local comparison runner requires the repository checkout. Synthetic success does not approve a release or change the default 96-tool catalogue.