Home/Blog/AI SEO
AI SEO

How to Connect Google Search Console to Claude via MCP (2026)

Rohit Sharma
Written byRohit Sharma
PublishedNov 4, 2026
Reading Time11 min read
CategoryAI SEO
Search performance dashboard with a rising clicks graph connected through a secure key connector to an AI chat window showing an insight card

Three ways to connect Google Search Console to Claude with MCP, from Node and Python servers to a web connector. Includes 15 copy-paste SEO prompts.

TL;DR: You can connect Google Search Console to Claude through an MCP server in about 15 to 20 minutes, giving Claude live access to your search data without 1,000-row CSV exports.

Google has no official Search Console MCP server, so you use a free community server with your own Google Cloud credentials.

We query our own SEOShouts property this way through a service account.

The catch: Claude still sends the returned data to Anthropic as part of the chat, so "local" does not mean private.

Connecting Google Search Console to Claude through MCP gives Claude live access to your search no CSV exports, no 1,000-row limit, no stale numbers.

There are three practical methods in November 2026.

Pick the one that fits from the table below, then follow that path.

MethodTools RequiredBest ForTime
Path 1: Node server (suganthan-gsc-mcp)Node.js 18+, Google Cloud projectMost users and agencies, 29 tools15 to 20 min
Path 2: Python server (mcp-gsc)uv, Google Cloud projectPython users, 20 tools15 min
Path 3: Remote connectorBrowser, plus a hosted server or third-party connectorUsing GSC in claude.ai on the web5 to 30 min

Every path needs Google Cloud credentials. The older "no Google Cloud needed" shortcut is gone: the popular npm servers removed their shared sign-in clients in 2026, so each user now creates their own.

The local paths also need an MCP client that runs local servers, such as Claude Desktop, Claude Code, Cursor or Codex.

Before You Start: Create Google Cloud Credentials

You do this once, and every path below uses it.

Step 1: Create a project and enable the API

Go to the Google Cloud Console and create a new project, for example "GSC MCP".

Open the API Library and enable the Google Search Console API.

Step 2: Choose OAuth or a service account

OAuth (personal use): go to Credentials, create an OAuth client ID, choose Desktop app and download the JSON file.

On first use a browser window asks you to sign in with the Google account that can see your properties.

Service account (agencies and automation): go to IAM and Admin, then Service Accounts, create one, then add a JSON key and download it.

Store the key file somewhere safe, because anyone with it can read your Search Console data.

Step 3: Give the service account access in Search Console

Skip this step if you use OAuth.

Open the key file and copy the client_email value.

In Search Console, open your property, go to Settings, then Users and permissions, and add that email as a user.

Full permission covers reporting; some write tools, such as sitemap submission, may need owner-level access.

Step 4: Note the exact property format

This is the most common setup mistake.

The property string in your config must match how the property appears in Search Console.

Search Console property typeUse this in config
example.com (domain property)sc-domain:example.com
https://example.com/ (URL prefix)https://example.com/

When in doubt, open the property selector in Search Console and copy the exact string.

Path 1: Node Server With 29 Tools (suganthan-gsc-mcp)

This server by Suganthan Mohanadasan is published on npm as suganthan-gsc-mcp.

Version 2.5 (August 2026) has 29 tools covering search analytics, content decay, keyword cannibalisation, CTR opportunities, multi-site dashboards, image search reports and optional URL submission.

It is read-only by default, and you opt in to write access during setup.

Best for: agencies managing several properties and anyone who wants ready-made analysis tools.

Step 1: Run the setup wizard

Open a terminal and run:

npx -y suganthan-gsc-mcp setup

The wizard asks for your OAuth client file or service account key and your property, then writes the configuration.

Step 2: Or add it to Claude Desktop by hand

Open Claude Desktop, go to Settings, then Developer, then Edit Config, and add this inside "mcpServers":

"gsc": {
  "command": "npx",
  "args": ["-y", "suganthan-gsc-mcp"],
  "env": {
    "GSC_KEY_FILE": "/path/to/service-account.json",
    "GSC_SITE_URL": "sc-domain:yourdomain.com"
  }
}

For OAuth instead, set GSC_AUTH_MODE to oauth and GSC_OAUTH_SECRETS_FILE to your client file.

For several properties, list them in GSC_SITE_URLS, separated by commas.

On Windows, use forward slashes in JSON paths, or escape each backslash as \\.

Step 3: Or add it to Claude Code

claude mcp add gsc -e GSC_KEY_FILE=/path/to/service-account.json -e GSC_SITE_URL=sc-domain:yourdomain.com -- npx -y suganthan-gsc-mcp

Run /mcp to confirm the server is connected.

Step 4: Restart and test

Quit Claude Desktop completely and reopen it.

In a new conversation, ask: "What were my top 10 pages by clicks in the last 28 days?"

If nothing comes back, check the config for JSON syntax errors and restart again.

Path 2: Python Server With 20 Tools (mcp-gsc)

The mcp-gsc server by Amin Foroutan is the most starred Search Console MCP project on GitHub, with about 2,100 stars in November 2026.

It now runs through uvx, so there is no cloning or virtual environment to manage.

Install uv first:

curl -LsSf https://astral.sh/uv/install.sh | sh

On Windows, use the PowerShell installer from the uv documentation.

Then add this to your Claude Desktop config:

"gscServer": {
  "command": "/FULL/PATH/TO/uvx",
  "args": ["mcp-search-console"],
  "env": {
    "GSC_OAUTH_CLIENT_SECRETS_FILE": "/full/path/to/client_secrets.json"
  }
}

For a service account, use GSC_CREDENTIALS_PATH instead.

Ask Claude to "call get_capabilities" to list all 20 tools and check the sign-in status.

If an older install crashes on start, update it: the MCP Python SDK 2.0 released in July 2026 broke earlier versions, and the current package pins a compatible SDK.

Path 3: Search Console in claude.ai on the Web

Local servers run on your computer, so they work in Claude Desktop and Claude Code but not in claude.ai in the browser.

For the web version you need a remote server that claude.ai can reach over HTTPS, added as a custom connector in Settings, then Connectors.

There are two ways to get one.

Host your own: the open source gsc-mcp-server package includes an HTTP mode you can deploy on a host such as Railway, Render or a VPS behind HTTPS.

It still uses your own Google Cloud credentials.

Use a third-party connector: services such as Windsor.ai and Adzviser connect to the Search Console API through OAuth and act as remote MCP servers.

Setup takes a few minutes, but your Search Console data passes through their servers, and pricing varies, so check their current plans.

5 Common Setup Errors and Fixes

No tools appear after restart. Your JSON config has a syntax error.

A missing comma or bracket breaks it without a clear message.

Paste the config into a JSON validator, fix it, save and restart.

Sign-in failed. Delete the cached token for your server and sign in again.

For service accounts, confirm the client_email in your key matches the user you added in Search Console.

404 error on domain properties. You used the wrong property format.

A domain property needs sc-domain:example.com, not https://example.com/.

Connected but no data returns. The account or service account has no access to that property.

Open Search Console as that account and check the property appears.

Windows paths not working. Use forward slashes, or escape every backslash as \\, in JSON config paths.

15 Copy-Paste Prompts for Your Search Console-Connected Claude

Once connected, these prompts give you immediate analysis.

Replace [domain] and date ranges as needed.

Quick wins

1. "Show pages on [domain] with over 500 impressions and CTR below 2% in the last 28 days, sorted by impressions."

2. "Which pages rank in positions 8 to 15 for queries with over 100 impressions?"

3. "Find queries where I rank in the top 3 but CTR is below average for that position."

Content decay

4. "Compare my top 20 pages for the last 28 days against the same period 90 days ago, and list those that dropped more than 20% in clicks."

5. "Which pages were in my top 10 by clicks 6 months ago but are outside the top 20 today?"

Cannibalisation

6. "Find queries where two or more of my pages compete, with URLs and average positions."

7. "Are there queries where both a blog post and a service page rank, and which URL should own the query?"

Indexing

8. "Inspect these URLs and return their index status and any issues: [paste list]."

9. "For my top 50 pages by impressions, which are not indexed, and why?"

Reporting

10. "Create a weekly summary: total clicks, impressions, average CTR, average position, and the top 5 pages and queries."

11. "Give me a prioritised list of 10 SEO actions for this week based on my Search Console data."

12. "Compare my best pages on mobile and desktop, and flag any heavily skewed to one device."

13. "Which pages have the most impressions but zero clicks, and what are the likely causes?"

14. "Is my average position over the last 90 days improving, stable or declining?"

15. "Which pages had strong CTR last quarter but dropped this quarter, as a table?"

For pages Claude flags as underperforming, run the specific URLs through the SEOShouts On-Page SEO Analyzer.

Search Console tells you which pages have problems.

The on-page analyser helps explain why.

What Is MCP and Why Does It Change How You Use Search Console?

MCP (Model Context Protocol) is an open standard that lets AI assistants connect to external tools and data through one common interface.

It is now governed by the Agentic AI Foundation under the Linux Foundation, and its July 2026 specification made remote servers simpler to run.

Before MCP, getting Search Console data into Claude meant exporting a CSV capped at 1,000 rows, uploading it and analysing a frozen snapshot.

With MCP, Claude queries the Search Console API directly, which returns up to 25,000 rows per request.

One important nuance: the MCP server passes Google's numbers through unchanged, but Claude's interpretation can still be wrong.

It might blame a traffic drop on an algorithm update when the data shows no such thing.

A simple guardrail is to tell Claude: "Base your analysis only on the Search Console data returned, and say you don't know rather than guess at causes the data doesn't show."

A note on URL submission: some servers include a tool for Google's Indexing API, but Google supports that API only for job posting and livestream pages, so don't rely on it for ordinary articles.

For how AI agents interact with websites more broadly, the WebMCP guide covers the browser-level protocol.

For how AI crawlers like GPTBot and ClaudeBot reach your pages in the first place, see the AI crawlers and internal links guide.

To compare Search Console with the other SEO data sources you can connect, see our overview of MCP servers for SEO.

Frequently Asked Questions

What is MCP and why does it matter for Google Search Console?

MCP (Model Context Protocol) lets AI assistants like Claude connect to live data sources.

For Search Console, it means Claude queries your actual search data instead of a CSV capped at 1,000 rows.

Does connecting Search Console to Claude expose my data to Anthropic?

Yes, the data Claude reads becomes part of your conversation.

A local server fetches it straight from Google on your machine, but the results it returns are sent to Anthropic so Claude can analyse them, under your Claude plan's data terms.

Third-party remote connectors also process the data on their own servers.

Use read-only access and avoid connecting properties you are not allowed to share.

Which method is best for agencies managing multiple sites?

Path 1 with a service account.

Add the service account to each client property and list them in GSC_SITE_URLS, then switch between properties without signing in again.

Can I use this with browser-based claude.ai?

Not with a local server.

Local servers work in Claude Desktop and Claude Code.

For claude.ai on the web, host a remote server over HTTPS or use a third-party connector, then add it under Settings, then Connectors.

What is the most common reason setup fails?

The wrong property format in the config.

A domain property (example.com in Search Console) needs sc-domain:example.com.

Using https://example.com/ for a domain property returns a 404.

Copy the exact string from the property selector rather than guessing.

Newsletter

Get More SEO Insights

Join 10,000+ marketers receiving weekly SEO strategies and tips directly in their inbox.

✓No spam, ever✓Unsubscribe anytime✓Expert SEO tips
Written by

Rohit Sharma

SEO Strategist & Founder at SEO Shouts

Results-driven SEO strategist with 13+ years of experience helping local, eCommerce, and B2B brands turn search visibility into qualified leads and revenue. Founder of SEOShouts and primary author behind its research, playbooks, and case notes.

No headings found

Continue Your SEO Journey

Discover more expert insights, proven strategies, and actionable tips to dominate search rankings.