What is Google Search Console MCP (mcp-gsc)?
mcp-gsc, by Amin Foroutan, lets an MCP client query Google Search Console: list properties, pull clicks, impressions, click-through rate and position by query, page, country or device, compare two periods, inspect up to 10 URLs at a time for index status, last crawl and rich results, and read sitemaps with their errors and warnings. It runs on your machine over stdio and signs in with your own Google OAuth client or a service account.
Google publishes no Search Console MCP server of its own, so this is the reference point for the category: about 2,200 GitHub stars, 283 forks and some 53,000 downloads of the mcp-search-console package from PyPI in the last month. It is MIT-licensed Python and reached version 0.4.2 on 6 November 2026. It does not call Google's Indexing API, so it cannot ask Google to crawl or index a page.
How to install Google Search Console MCP (mcp-gsc)
uvx mcp-search-consoleIs Google Search Console MCP (mcp-gsc) worth it?
The useful part of mcp-gsc is that it turns the questions an SEO already asks in the Search Console interface into tool calls a model can chain together. Which queries lost clicks between two 28-day periods, which pages rank on page two with high impressions, whether ten important URLs are actually indexed and when Google last crawled them. Advanced analytics pages through results beyond the API's 25,000-row limit per request, and a batch inspection checks up to 10 URLs per call.
The permission model is now the right one. Version 0.4.2 requests only Google's webmasters.readonly scope unless GSC_ALLOW_DESTRUCTIVE=true is set, in which case it asks for full write access and enables adding and deleting properties and deleting sitemaps. A submit_sitemap tool is still listed in the default set, but a read-only token gives Google no grounds to accept it. Before that release, the server asked for broader access than its default tools needed, and a failed token refresh could delete the saved token; both are fixed in the changelog.
The catch: setup is the hardest part, and it is Google's doing. You create a Google Cloud project, enable the Search Console API, configure an OAuth consent screen and download a client secrets file, or create a service account and add it to each property with full access. The README's own video walkthrough exists for a reason. Python 3.11 is required, and under Claude Desktop an older Python fails silently with no tools and no log. Google's quotas also bind on large sites: URL inspection is limited to 2,000 calls a day and 600 a minute per property, so a model cannot audit a 50,000-page site in one sitting.
It does less than some newer rivals. There is no Indexing API, no Bing Webmaster Tools and no GA4 in the open-source server; the author steers those features to a paid hosted version, and the README carries banners for it. Servers such as FlorianBruniaux's google-search-console-mcp add indexing submissions and Bing, at the cost of a wider attack surface.
Our take: the read-only default is the feature that matters most for a tool that holds credentials to your search data, and this one now gets it right. Most SEO teams need analysis rather than writes, which is exactly the trade it makes.
Where it earns its place
An in-house SEO at a 2,000-page e-commerce site
Find out why organic clicks fell last month
Ask for a comparison of the last 28 days against the previous 28 by page and query, then inspect the ten biggest losers for index and canonical problems. A morning of spreadsheet work becomes a few prompts; the API itself is free.
A freelance SEO consultant with eight client properties
Produce a monthly striking-distance report for each client
A service account added to each property lets the agent pull queries ranking in positions 8 to 20 with high impressions and draft recommendations. Keep the default read-only scope so a mistaken prompt cannot delete a client's sitemap.
A developer who has just migrated a blog to a new framework
Check that key pages are indexed and their sitemaps parse
Run batch inspection on the most important URLs and read the sitemap error counts. Inspection quota is 2,000 calls a day per property, enough for a small site's spot checks.
Before you commit
- Setup
- Create a Google Cloud project, enable the Search Console API and either an OAuth desktop client or a service account added to each property; then point the client at uvx mcp-search-console with GSC_OAUTH_CLIENT_SECRETS_FILE or GSC_CREDENTIALS_PATH. The first OAuth run opens a browser to sign in. Expect 20 to 40 minutes the first time.
- Integrations
- About 20 tools for properties, search analytics, period comparison, URL inspection and sitemaps, plus Claude Code and Cursor plugin manifests and a Dockerfile. Stdio by default; an SSE mode exists for remote use.
- Data and privacy
- Search Console data flows from Google to your machine and then to your model provider. The OAuth token is saved as token.json in your user config folder. The SSE mode turns off the MCP SDK's DNS-rebinding protection when bound to a remote host, so keep it behind your own authentication.
- Cost at scale
- Free; the Search Console API has no charge. The limit at scale is quota, not money: 2,000 URL inspections a day per property and 1,200 search analytics queries a minute per site.
- Limits
- Search analytics returns at most 25,000 rows per request (the server pages beyond it). No Indexing API, so no crawl requests. Eight open issues and pull requests. Service accounts need full access on each property for inspection.
What’s good
- Read-only Google scope by default since 0.4.2
- Period comparison and pagination past 25,000 rows
- Batch URL inspection of up to 10 URLs
- OAuth or service account; runs locally over stdio
What’s not
- Google Cloud project and OAuth setup before the first query
- No Indexing API, Bing or GA4 in the free server
- Python 3.11+ required; older versions fail silently in Claude Desktop
- README promotes the author's paid hosted version
Best for
SEOs and site owners who want an assistant to analyse their own Search Console performance and indexing data without giving it write access.
Google Search Console MCP (mcp-gsc) alternatives
Google Search Console MCP (mcp-gsc): common questions
- Is Google Search Console MCP (mcp-gsc) worth it?
- For an SEO who wants to ask questions of their own Search Console data, this is the sensible default, and since 0.4.2 it can no longer change anything unless you tell it to. We rate it 4.2 out of 5.
- How much does Google Search Console MCP (mcp-gsc) cost?
- Free and open source (MIT). The Search Console API costs nothing; the author also sells a separate hosted version with GA4 and keyword tools, with prices on its own site.
- What are the downsides of Google Search Console MCP (mcp-gsc)?
- Google Cloud project and OAuth setup before the first query; no Indexing API, Bing or GA4 in the free server; python 3.11+ required; older versions fail silently in Claude Desktop; rEADME promotes the author's paid hosted version.
- Who is Google Search Console MCP (mcp-gsc) best for?
- SEOs and site owners who want an assistant to analyse their own Search Console performance and indexing data without giving it write access.
- How much does Google Search Console MCP (mcp-gsc) cost for a team or in production?
- Free; the Search Console API has no charge. The limit at scale is quota, not money: 2,000 URL inspections a day per property and 1,200 search analytics queries a minute per site.
- Is Google Search Console MCP (mcp-gsc) safe for company data?
- Search Console data flows from Google to your machine and then to your model provider. The OAuth token is saved as token.json in your user config folder. The SSE mode turns off the MCP SDK's DNS-rebinding protection when bound to a remote host, so keep it behind your own authentication.
- What are the best Google Search Console MCP (mcp-gsc) alternatives?
- The closest alternatives we have reviewed are Firecrawl MCP and Google Workspace MCP servers.
Answers checked Nov 13, 2026.
Make Google Search Console MCP (mcp-gsc)? Show this rating on your site
