# Google Search Console MCP: 7 Ways to Connect It to Claude

**Last updated:** 2026-10-10  
**Source:** https://www.lawrencehitches.com/google-search-console-mcp/

---

**A Google Search Console MCP connects Claude to your Search Console data, so you can ask about clicks, impressions, positions and queries in plain English. The easiest way in is a no-code connector like Windsor.ai or Porter Metrics: sign in with Google, add the connector in Claude, done.** There is no official Google server for Search Console, so every route below is third-party. This guide ranks them easiest first, compares the seven developer servers, then gives you the six questions to ask once it is connected and the workflows I run on this site.

MCP-GSC is also one of the [six Claude SEO plugins worth installing](https://www.lawrencehitches.com/claude-seo-plugins/#which-plugins), alongside Firecrawl, OpenSEO and Google Trends, if you want the whole toolbelt rather than one server.

## What is the easiest way to connect Google Search Console to Claude?

**The easiest way to connect Google Search Console to Claude is a no-code connector: Windsor.ai has a free plan, and Porter Metrics is a free connector you add with one URL.** Claude has no built-in Search Console switch, so you add it as a connector. Pick one route, give it read-only access, and check any change Claude suggests before it goes live. Free Claude plans can add one custom connector.

| Way to connect | Cost | Code needed | Best for |
|---|---|---|---|
| 1. Windsor.ai | Free plan (one data source) | None | Beginners, fastest start |
| 2. Porter Metrics | Free connector | None, paste one URL | Beginners who want a custom connector |
| 3. Supermetrics | Paid, 14-day free trial | None | Marketers already on Supermetrics |
| 4. Claude Desktop + gsc-mcp-server | Free, open source | One terminal command | Free and local, no third party in the middle |
| 5. Composio | Free developer tier | None, paste one URL | Connecting many apps through one URL |
| 6. Coupler.io | Free trial, then paid | None | Scheduled weekly reports, not live questions |
| 7. CSV export | Always free | None | Every plan, nothing to install |

### 1. Windsor.ai: no code, free plan

Go to onboard.windsor.ai, choose Google Search Console, sign in with Google and pick your site. In Claude, open Customize, then Connectors, search for Windsor and click Add. Sign in to Windsor when Claude asks. The free plan covers one data source, which is all you need for Search Console.

### 2. Porter Metrics: free connector, paste one URL

Log in to Porter with Google, give it read-only access to Search Console and pick your site. In Claude, open Customize, then Connectors, click + Add, then Add custom connector. Name it Porter, paste `https://mcp.portermetrics.com/mcp`, then sign in with the same email.

### 3. Supermetrics: in Claude's directory, paid

Open Claude's connector directory, find Supermetrics Marketing Analytics and click Add. Sign in to a Supermetrics account (paid, with a 14-day free trial), then connect Search Console inside Supermetrics.

### 4. Claude Desktop: free and open source

Run `npx gsc-mcp-server --setup` in your terminal. A browser opens, and you sign in with Google. That is the npm release (1.1.0, February 2026). The unreleased v2 on GitHub removes the shared Google sign-in, so if you install from source you will need your own Google Cloud credentials first (a service account for remote setups, or a client ID and secret for the setup command). Then open Claude Desktop, go to Settings, Developer, Edit Config, paste the block below, save and restart Claude.

```
"mcpServers": {
  "gsc-mcp-server": {
    "command": "npx",
    "args": ["-y", "gsc-mcp-server"]
  }
}
```

Several unrelated packages have used this name, so check the repo (github.com/sofianbettayeb/gsc-mcp-server) before you run it.

### 5 to 7. Composio, Coupler.io, or no connector at all

**Composio** works like Porter: add a custom connector with `https://connect.composio.dev/mcp`, sign in to Composio, then ask Claude to connect Google Search Console and approve the Google sign-in. **Coupler.io** copies your Search Console data on a schedule into a place Claude can read, so it suits weekly reports more than live questions. And with **no connector at all**, open Performance in Search Console, set the dates, click Export, choose CSV and upload the file to Claude. That works on every plan.

If you write code and want more control than a connector gives you, the developer servers are the next step.

## Which Google Search Console MCP server should you use?

**Seven credible open servers exist for developers, and the right one depends on your client and whether you want plain data access or opinionated analysis.** All seven verified live on GitHub as of 18 October 2026:

| Server | Install | Distinct strength |
|---|---|---|
| AminForou/mcp-gsc | uvx (no clone, no venv) | Smoothest zero-install path; broad client support (Claude Desktop, Cursor, Gemini CLI) |
| suganthan-gsc-mcp (npm) | npx | 20 built-in analysis tools (quick wins, cannibalisation, decay) plus guardrail prompts against hallucinated numbers |
| ahonn/mcp-server-gsc | npm | Long-standing, minimal, privacy-first local run |
| acamolese/google-search-console-mcp | local | Read-only OAuth by design plus brandable HTML audit reports |
| ncosentino/google-search-console-mcp | local | Deep row limits (up to 50,000 rows per query) for big-site pulls |
| saurabhsharma2u/search-console-mcp | local | GSC, Bing Webmaster and GA4 behind ONE tool set, with the maths done server-side |
| garethcull/search-console-mcp | Flask, self-host | Deployable to Cloud Run for team or agency infrastructure |

My picks: **uvx users start with AminForou's** for the frictionless install; **analysis-first users start with Suganthan's** for the built-in tool set; **agencies** look at the Flask or read-only options for infrastructure and safety. All of them need the same one-time Google authorisation, which is the actual setup step that matters.

## How does setup work?

**Setup is a one-time Google authorisation plus one config entry, roughly 15 minutes.** The shape is identical across servers: create (or reuse) a Google Cloud project, enable the Search Console API, create OAuth credentials, run the server once to complete the browser sign-in, then add it to your MCP client config. Service-account auth is the alternative for agencies: more steps, but tokens do not expire per-user and it suits automation. After that, the server appears in Claude Code as a set of callable tools, ready for your first questions.

## What should you ask Claude once Search Console is connected?

**Ask Claude these six questions first: they cover lost clicks, ignored impressions, titles, near-page-one pages, cannibalisation and your biggest fixes.** Paste each one as written. They work with any of the routes above, including a CSV upload.

1. **Which pages lost the most clicks?** "Using my Search Console data, which pages lost the most clicks in the last 28 days compared with the 28 days before? Show each page, clicks before, clicks after, and the change. Then tell me the likely reason for the top 3."
2. **Which searches get seen but not clicked?** "Using my Search Console data from the last 3 months, list the queries with the most impressions and a click-through rate under 2%. For each, show the page that ranks, its average position, and one reason people might not be clicking."
3. **Rewrite a title in the words people type.** "Here is a page: [URL]. Using my Search Console data, list the queries this page gets impressions for. Then rewrite its title tag in the words people actually type, under 60 characters. Give me 3 options."
4. **Which pages sit just off page one?** "Using my Search Console data from the last 3 months, list the pages with an average position between 8 and 15 and at least 100 impressions. For each, tell me the one change most likely to move it up."
5. **Which two pages are fighting for the same search?** "Using my Search Console data, find queries where two or more of my pages get impressions. Show the query, each page, and their clicks. Tell me which page should own each query and what to do with the other one."
6. **Give me my 5 biggest fixes.** "Using everything in my Search Console data, give me my 5 biggest fixes, ranked by how many clicks each could win back or add. Keep each fix to one line, and say which page it's for."

Once those answers come back, the real value is in the workflows you run on top of them.

## What do you actually DO with it? (the workflows that earn their keep)

**The value is not data access, it is that analysis happens where the follow-up questions are.** These are real workflows from this site, with what they caught:

- **The vanity-metric autopsy.** Ask for daily impressions, clicks and position across 90 days, then ask what changed on the spike date. On this site that surfaced a June impression spike (3.6x overnight, position improving from 48 to 21) that added about 96,000 impressions and no clicks: machine-generated operator queries, not demand. The follow-up question ("which queries drove the spike, and what is their CTR?") is exactly the step a dashboard never invites you to take.
- **The pre-edit query-profile check.** Before touching any title, pull every query the page ranks for. That check stopped me rewriting a tool page title that was holding position 9.75 for its exact-match term and position 3 for a branded variant. The rewrite would have chased a 60-impression keyword and risked both.
- **Striking-distance triage with a noise filter.** Positions 4 to 10 sorted by impressions, MINUS the junk: filter out anything containing site operators, after: dates, or URL strings. On this site the unfiltered list was 698 rows; the real list was 343. The filter is one sentence in a prompt.
- **Cannibalisation checks on demand.** "Which queries have two of my pages competing?" caught two pages splitting 107 impressions for one term, and the data picked the survivor (19 positions apart).

## How do you stop the model inventing numbers?

**Ground every claim in a tool call, and make the model show the pull.** Three habits that hold up: ask for the raw rows before the interpretation; never accept a number that did not come back from a tool call in the same conversation; and re-pull anything surprising with an explicit date range, since default windows differ between tools. Some servers ship guardrail prompts for exactly this, and they are worth keeping on. The same discipline applies to the other half of the stack, your Analytics data.

## FAQ

### Is there an official Google MCP server for Search Console?

No. Google ships an official MCP server for [Analytics](https://www.lawrencehitches.com/google-analytics-mcp/), but not for Search Console. Every route on this page is third-party, which is why comparing the options matters.

### Can I connect Search Console to Claude for free?

Yes. Windsor.ai's free plan, the Porter Metrics connector, the open-source Claude Desktop setup and a plain CSV export are all free. Free Claude plans can add one custom connector.

### Can a Search Console MCP server change my site or my GSC settings?

Only if you grant it write scopes. Several servers are read-only by design, and read-only is the right default for analysis work. Check the OAuth scopes on the consent screen before approving.

### Does this work with Claude Code specifically?

Yes. Claude Code supports MCP servers natively via its config, and every server in the table above speaks the same protocol. The same config also works in Claude Desktop and Cursor.

### How much Search Console history can I query?

Whatever the GSC API holds, which is 16 months. For longer memory, schedule exports, or pair the server with a warehouse; the MCP layer does not extend Google's retention.

## Sources

- AminForou/mcp-gsc: https://github.com/AminForou/mcp-gsc
- Suganthan Mohanadasan, GSC MCP setup guide: https://suganthan.com/blog/google-search-console-mcp-server/
- ahonn/mcp-server-gsc: https://github.com/ahonn/mcp-server-gsc
- acamolese/google-search-console-mcp: https://github.com/acamolese/google-search-console-mcp
- ncosentino/google-search-console-mcp: https://github.com/ncosentino/google-search-console-mcp
- saurabhsharma2u/search-console-mcp: https://github.com/saurabhsharma2u/search-console-mcp
- garethcull/search-console-mcp: https://github.com/garethcull/search-console-mcp
- Windsor.ai, sending Search Console data to Claude: https://windsor.ai/how-to-send-google-search-console-data-to-claude/
- Porter Metrics, Claude MCP tutorial: https://portermetrics.com/en/tutorial/ai/claude-mcp/
- Supermetrics connector in Claude's directory: https://claude.com/connectors/supermetrics
- sofianbettayeb/gsc-mcp-server: https://github.com/sofianbettayeb/gsc-mcp-server
- Composio, Google Search Console toolkit: https://composio.dev/toolkits/google_search_console/framework/claude-cowork
- Coupler.io, Search Console to Claude: https://www.coupler.io/claude-integrations/google-search-console-to-claude

---

*Lawrence Hitches is an AI SEO consultant in Melbourne and Chief of Staff at StudioHawk, Australia's most awarded SEO agency. Agent edition of this site: https://www.lawrencehitches.com/md/index.md*