What mcp-gsc gives you
20 tools: search analytics with filters and period comparisons, URL inspection for up to 10 URLs at a time, sitemaps and property management. It talks to Google with your own credentials, so your data goes from Google to your machine and nowhere else. It’s actively maintained, with a release in November 2026.
What you need
- A verified property in Google Search Console
- A Google Cloud account, which is free
uv, which runs the server without you managing Python- An MCP client on your machine: Claude Desktop, Cursor, Codex CLI, Gemini CLI or Antigravity
Step 1: Create Google credentials
- Create or pick a project in the Google Cloud Console.
- Enable the Search Console API for it.
- Configure the OAuth consent screen.
- Create an OAuth client ID of type Desktop app, download the JSON file and save it somewhere permanent.
Step 2: Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
Then run which uvx and copy the full path it prints. The config needs it, because desktop apps don’t read your shell’s PATH.
Step 3: Add mcp-gsc to your client
For Claude Desktop, edit claude_desktop_config.json:
{
"mcpServers": {
"gscServer": {
"command": "/FULL/PATH/TO/uvx",
"args": ["mcp-search-console"],
"env": {
"GSC_OAUTH_CLIENT_SECRETS_FILE": "/full/path/to/client_secrets.json"
}
}
}
}
Cursor takes the same block in ~/.cursor/mcp.json, or you can install mcp-search-console from the Cursor Marketplace. Codex CLI takes the same command and environment variable in ~/.codex/config.toml. Check the README before you paste: open-source projects change faster than guides.
Step 4: Restart and sign in
Quit the app completely. Closing the window isn’t enough. On first use a browser window opens for Google sign-in, and the token is saved after that. Ask your assistant to “call get_capabilities” to confirm it’s connected and see every tool.
Use a service account instead
Better for automation or a shared setup. Create a service account and a JSON key in Google Cloud, then add the service account’s email to your Search Console property as a user with Full access. In the config, set GSC_CREDENTIALS_PATH to the key’s path and GSC_SKIP_OAUTH to "true".
Write tools are off by default
Adding or deleting properties and deleting sitemaps stay disabled unless you set GSC_ALLOW_DESTRUCTIVE to "true". I’d leave them off. Nothing in day-to-day analysis needs them.
Running mcp-gsc on a server
It can also run over HTTP (SSE) or in Docker, which is how you’d put it on a public URL for claude.ai. That means hosting and securing it yourself, and at that point a hosted server is less work.
If mcp-gsc doesn’t start
spawn uvx ENOENT: the app can’t find uvx. Use the full path fromwhich uvxin the config.- No tools show up after cloning the repo: that route needs Python 3.11 or newer. The uvx install handles Python for you.
- It works in Claude Desktop but not on claude.ai: expected. It runs on your machine.
Other open-source Search Console MCP servers
- search-console-mcp (saurabhsharma2u on GitHub, MIT): adds Bing Webmaster Tools, GA4 and AdSense, and signs you in with
npx search-console-mcp setup, so there’s no Google Cloud project for Search Console. - mcp-server-gsc (ahonn on GitHub): another option, but GitHub shows no license file for it, so check with the author before using it at work.
Is self-hosting worth it?
It is if you’re a developer, you want your data to stay in your own Google project, and Search Console data is all you need.
It isn’t if you want this working in five minutes, or on claude.ai or ChatGPT’s website. A hosted server skips every step above. Search Console MCP has a free plan for raw data, and LinkGPT adds keyword research, AI visibility and ranked fixes on top of Search Console. All six are compared in the Google Search Console MCP server list.