Google Search Console MCP: How to Connect It to Claude and ChatGPT, and Use It for Every Client (2026)

As of October 2026 Google doesn't publish an official Search Console MCP server, so every Google Search Console MCP is third-party code built on Google's public Search Console API. The usual setup is an open-source server on your own computer: a Google Cloud project with the Search Console API turned on, a service account or Google login that can see your properties, and one config entry in Claude.

Below: what any Search Console MCP can read, the exact setup for the most-used open-source server, and what changes with a dozen clients.

Is there an official Google Search Console MCP server?

No. An MCP server is an address or small program you add to Claude or ChatGPT so it can read another system. Google runs remote MCP servers for many of its products, but Search Console isn't on Google Cloud's list of supported products (85 entries in October 2026), and we found no Search Console MCP repository in Google's GitHub organizations. Google does publish one for Analytics (see our Google Analytics MCP guide).

The real options:

• An open-source server you run yourself. The most used is mcp-gsc by Amin Foroutan (about 2,000 GitHub stars in October 2026, Python, MIT licence). A smaller, read-only one is mcp-server-gsc by ahonn (about 280 stars, Node.js).
• A hosted server from a third party. A company runs it and you sign in with Google.

What a Search Console MCP server can and can't read

Every server is limited by the same API:

• Four parts: the API has Search Analytics (the Performance report), Sitemaps (list, get, submit, delete), Sites (list, add, remove properties) and URL Inspection, which reports a URL's status in Google's index. The API can't request indexing.
• Read or write is set by the scope. webmasters.readonly is enough to read performance data and inspect URLs; submitting a sitemap needs the full webmasters scope.
• Metrics: clicks, impressions, CTR and average position, by query, page, country, device, search appearance, date or hour, for web, image, video, news, Discover or Google News (query reference).
• Delay: Google says data is normally available after 2 to 3 days. Asking for dataState: all adds fresher, still-incomplete days.
• History: 16 months.
• Rows: up to 25,000 per request, and at most 50,000 rows per day per site per search type. Anonymized queries (searched by fewer than a few dozen people) never appear as rows but are counted in totals, so query rows rarely add up to the total.
• Quotas: URL Inspection is capped at 2,000 a day per site (usage limits).
• Property types: a Domain property covers every subdomain and protocol and is verified by DNS; a URL-prefix property covers only URLs starting with that exact prefix. The API writes them as sc-domain:example.com and https://www.example.com/ (with the trailing slash).

No Search Console MCP can give you backlinks, competitors' rankings or search volume.

How to set up a Search Console MCP server in Claude, step by step

This uses mcp-gsc (version 0.4.1) with a service account, so clients add one email instead of you signing in to their Google accounts. Steps are from its README as of October 2026.

You need: a Google Cloud project, Owner access to each Search Console property (only owners can add users), and uv, which runs the server without installing Python by hand.

Step 1: Turn on the Search Console API

In your Google Cloud project, enable the Google Search Console API.

Step 2: Create a service account and key

Go to APIs & Services, Credentials, Create Credentials, Service Account. On its Keys tab, choose Add Key, Create new key, JSON, and save the file somewhere permanent.

Step 3: Add the service account to the property

Copy the service account's email (it ends in .iam.gserviceaccount.com). In Search Console, open the property, then Settings, Users and permissions, Add user, paste the email and choose a permission. The README says Full; Restricted (view rights on most data) is safer if you only want reports.

Step 4: Add it to Claude Desktop

Open Settings, Developer, Edit Config, which opens claude_desktop_config.json (Anthropic's guide), and add the README's service-account block. Use the full path from which uvx; Claude Desktop doesn't read your shell's PATH.

{
  "mcpServers": {
    "gscServer": {
      "command": "/FULL/PATH/TO/uvx",
      "args": ["mcp-search-console"],
      "env": {
        "GSC_CREDENTIALS_PATH": "/full/path/to/service_account.json",
        "GSC_SKIP_OAUTH": "true"
      }
    }
  }
}

Quit Claude Desktop completely and reopen it. In Claude Code, the same server is one command (Claude Code's MCP docs):

claude mcp add --transport stdio --scope user \
  --env GSC_CREDENTIALS_PATH=/full/path/to/service_account.json \
  --env GSC_SKIP_OAUTH=true \
  gsc -- uvx mcp-search-console

Prefer your own Google login? Create a Desktop-app OAuth client, download its JSON and set GSC_OAUTH_CLIENT_SECRETS_FILE to its path instead; you sign in in the browser on first use.

Step 5: Try it

Ask "List my GSC properties", then: "Compare clicks for sc-domain:example.com over the last 28 days against the 28 before, and list the queries that lost the most." If nothing shows up, ask Claude to call get_capabilities, which reports the login status.

Claude on the web and ChatGPT

Both connect to servers by address, not to a program on your computer (OpenAI's MCP docs). mcp-gsc can run as a web server or in Docker; mcp-server-gsc has a Cloudflare Workers version with a bearer token. If you host it, put sign-in in front of it, or anyone with the address can read your clients' data. In Claude you then add it under Customize, Connectors, Add custom connector, where you can set an OAuth client or request headers. Which ChatGPT plans can add custom servers has changed several times, so check OpenAI's current help page.

What to check before connecting a client's property

• Read or write: mcp-gsc asks Google for the full webmasters scope and can submit sitemaps by default; adding and removing properties and deleting sitemaps stay off unless you set GSC_ALLOW_DESTRUCTIVE=true. mcp-server-gsc asks only for webmasters.readonly. Either way, a Restricted user in Search Console can only look.
• Who holds the key: with a self-hosted server, it's the JSON file on each laptop. With a hosted server, it's the vendor, for every client you connect.
• Who sees what: every property its Google identity can see, with no per-teammate filter.
• Google's terms: using the API means accepting the Google APIs Terms of Service; check what your client agreements say about where their data goes.

Using a Search Console MCP across many client properties

Search Console has no manager account like Google Ads. Access is per property: each client's owner adds your service account email as a user, like adding a colleague. Then the Sites list returns every property it was added to with its permission level, so Claude can pick a client by domain. Each property allows up to 100 non-owner users.

Say you look after 14 clients' SEO and each has added your service account. On Monday you ask: "Which clients lost the most non-brand clicks in the last 28 days?" Claude lists the properties, guesses each client's brand terms, then compares two periods 14 times. That works, but it's a long chain of tool calls, and how Claude reads "non-brand" can change from week to week.

• Teammates: everyone with the key file sees all 14 clients. Keeping a freelancer to three means a second service account added only to those three.
• Domain or URL-prefix: add the service account to the property you report from; some clients only verified a URL-prefix one.
• Clients wanting their own access set up the same thing themselves.
• Search is one source. GA4, ads and the CRM need separate servers. On the safety side of ad accounts, see connecting Claude to client Meta and Google Ads safely.

Where ExpertRelay fits

ExpertRelay gives an agency its own hosted MCP server for the team, plus one per client if you want it. You add the address in Claude (Customize, Connectors, Add custom connector) or ChatGPT (Settings, Connectors, Create, where the plan allows), and each person signs in with their own login.

For Search Console there's no Cloud project: the agency gets its own Google service-account email, and each client adds it as a user on their property. ExpertRelay then reads clicks, impressions, click-through rate and average position, per day and for each search query: daily figures go back 30 days, and query totals compare the last 14 days with the 14 before. It never submits sitemaps, requests indexing or edits anything, and doesn't read backlinks, crawl errors or competitors' rankings.

The Monday question becomes one call: compare_clients puts every client side by side, biggest drops first, next to their GA4, Google Ads, Business Profile and other sources; run_for_client runs one of your playbooks on one client's data. Team members reach only the clients they work with; owners and admins see all. Every call is logged. More on this for SEO teams on ExpertRelay for SEO agencies.

Open-source server vs. hosted server vs. ExpertRelay

• Open-source Search Console MCP server: free, any breakdown the API allows, URL inspection, and sitemaps if you allow it. Limitation: runs on each person's computer with a key file, needs self-hosting for Claude on the web or ChatGPT, and can't keep a teammate to some clients.
• A third-party hosted Search Console server: a company runs it and you sign in with Google, so it works in the web apps. Limitation: another company holds access to your clients' properties, plans and limits vary, and many are built for one site owner rather than an agency's client list.
• ExpertRelay: the agency's own server with every client's Search Console, GA4 and ad platforms side by side, access per teammate, a server per client if you want it, and your playbooks. $99 a month for the agency (whole team, up to 25 clients), $49 a month for each client who gets their own AI, 14 days free. Limitation: daily rows and query totals, not raw API access, so no URL inspection, page-level breakdowns or custom regex filters; it can't change anything; you can't add your own code or other MCP servers; ChatGPT custom connectors depend on the ChatGPT plan.

When an open-source server is enough

• One site, or a few you check yourself. It costs nothing to run.
• You need page-level or URL inspection work. Cannibalization checks and indexing audits need the full API.
• You want it to submit sitemaps. mcp-gsc can, with the full scope and a permission level above Restricted.

FAQ

Does Google have an official Search Console MCP server?

Not as of October 2026. Search Console isn't on Google's list of products with MCP servers, and every GSC MCP server is third-party code using the public Search Console API.

What is the best Google Search Console MCP server?

The most used open-source one is mcp-gsc, with about 2,000 GitHub stars. If you want one that can only read, mcp-server-gsc requests the read-only scope.

Can a Search Console MCP request indexing?

No. The API can inspect a URL's index status but can't request indexing. Some servers can submit sitemaps with write access.

Does the GSC MCP work with ChatGPT?

Only through a server hosted at a web address, either one you host or a third party's, and only on ChatGPT plans that allow custom MCP servers.

How do I give an MCP server access to a client's Search Console?

The property's owner goes to Settings, Users and permissions, Add user, and adds the service account email. Restricted gives view rights on most data, which covers reporting.

If your team wants every client's Search Console, GA4 and ad numbers in Claude or ChatGPT without a key file on each laptop, see the agency's own MCP server or ExpertRelay for agencies.