If you want to self-host instead, community servers such as ahonn/mcp-server-gsc are open code you run yourself, authorised through a Google Cloud service account. The comparison below covers when each one fits.
What the server gives Claude
Search Console already answers the questions people bring to Claude, but getting those answers usually means exporting a CSV and sorting it by hand. With the server connected, you ask in plain words and Claude picks the tool: “which queries are sitting on page two with real impressions?”, “which of my pages compete for the same search?”, “what did I publish that Google has never shown anyone?”.
Cituna’s tools do the sorting for you. Each one returns an answer rather than raw rows: the striking-distance list comes ranked, zero-click pages come with the cause, and cannibalised queries come with the page Google prefers. The raw query tool is there for anything else.
The other half is the AI side. Every tracked buyer question is asked of seven engines daily, and the server exposes those answers too. One built-in workflow, gsc-to-ai-gap, takes the queries where Google already shows you and checks which ones the AI engines never cite you for, which gives you a ranked list for answer-engine work. A Search Console-only server cannot make that join, because it has no AI answers to compare against. The Google Search Console integration page covers what the joined view looks like in the app.
The nine Search Console tools
| Tool | The question it answers | Defaults and limits |
|---|---|---|
gsc_status | Is Search Console connected, and which properties can I query? | Lists every connected Google account and its properties |
gsc_overview | How is search traffic doing? | Totals, top queries and pages, country, device and a daily series; up to 90 days |
gsc_query | Anything else, by query, page, country, device, search appearance or date | Up to 25,000 rows; up to 480 days |
gsc_striking_distance | Which queries are close enough to page one to move this week? | Positions 11 to 60 by default, ranked by a priority score |
gsc_zero_click_pages | Which pages get impressions and no clicks, and why? | Split into branded, snippet, page-two and ranking causes; 25-impression floor by default |
gsc_query_mix | What kind of demand do I actually earn? | Brand, competitor, persona, commercial, tool, how-to and category buckets |
gsc_coverage_gap | Which published pages has Google never shown? | Reads your sitemap; 90-day window by default |
gsc_cannibalisation | Where do two of my own pages compete for one query? | Every competing page, its position and the impression split |
gsc_ai_searches | Which searches are worded like AI prompts rather than keywords? | A floor on AI use, not Google’s AI Overviews numbers |
Search Console data lags by about two days, so each window ends roughly two days before today. Google also withholds rare queries, which is why a page’s named-query impressions can come in below its total.
Hosted or self-hosted: which GSC MCP server to install
The results for this search are mostly community servers. They are useful, and the right choice when you only want Search Console in a chat and are happy to run the code yourself. ahonn/mcp-server-gsc is a typical one: you create a Google Cloud project, enable the Search Console API, create a service account and download its JSON key, add the service account’s email as a user on your Search Console property, then point npx -y mcp-server-gsc at the key through the GOOGLE_APPLICATION_CREDENTIALS variable. Others, such as AminForou/mcp-gsc, are Python scripts you run the same way.
| Cituna (hosted) | A self-hosted community server (e.g. ahonn/mcp-server-gsc) | |
|---|---|---|
| Where it runs | Cituna’s servers, at one URL | Your machine, launched by the client |
| Google sign-in | Browser consent once, in the Cituna app | A service account JSON key you create and store |
| Google Cloud project | Not needed | Needed, with the Search Console API enabled |
| Works in Claude on the web and mobile | Yes | No (needs a local process) |
| Search Console scope | Read-only (webmasters.readonly) | Whatever the service account is granted |
| Several Google accounts | Yes, one workspace can connect several | Each property must add the service account as a user |
| Ready-made analyses | Nine tools, including striking distance, zero-click causes and cannibalisation | Depends on the server |
| AI engine answers | Seven engines, asked daily, in the same connection | None |
| Cost | Included in every paid plan, from $49 a month | No plan to buy; you host it |
Pick a self-hosted server if Search Console is all you need and you are comfortable running Node or Python and holding a Google key. Pick Cituna if you want a connector that also works in Claude on the web, with no key on your laptop, or if the question you keep asking is why AI engines cite someone else for searches you already rank for.
Connect it in three steps
- Connect Search Console in the app. In Cituna, open Integrations and connect your Google account. Google asks for read-only Search Console access. Choose the account where your domain is a verified property; the Search Console connection guide covers the case where it is not.
- Add the connector to Claude. In Claude, open Customize, then Connectors, add a custom connector and paste
https://mcp.cituna.com/mcp. - Approve it. Sign in to Cituna and approve the connection. Untick “Also allow changes” if you want the connection to stay read-only.
Then ask Claude to run gsc_status; it lists each connected account and the properties it can read. The same URL works in ChatGPT, Cursor, VS Code, Windsurf and other MCP clients. For scripts and CI, the identical server is on npm as cituna-mcp, which needs Node 20 or later and an API key. For the questions worth asking once it is connected, see how to query Search Console from Claude. For every tool, Search Console and beyond, see the full MCP server page.
How your Google data is handled
The MCP server holds no Google credentials. Your Search Console refresh token stays on Cituna’s backend, which mints a short-lived Google access token for each request. Cituna asks Google for the webmasters.readonly scope only, so nothing it does can change your Search Console settings or properties. Access tokens for the connector last 24 hours, and disconnecting Search Console in the app cuts the server off straight away.
What it cannot tell you
- Google’s own AI Overviews and AI Mode impression totals. The Search Analytics API has no filter for them.
gsc_ai_searchespicks out searches worded like prompts, which is a floor on AI use, not Google’s figure. - Pages on a property your Google account has not verified. If the account you connect has never verified the domain, the tools connect fine and come back empty.
- The last two days. Search Console reports with a lag of about two days, so every window ends before them.
Which plan you need
- Starter, $49 a month: all nine Search Console tools and the other read tools (every one of the 22 except
list_keywords, which needs Pro), so Claude can read your AI answers, audits and fix queue next to your Google data. On Starter the connection is read-only. - Pro, $119 a month: adds the nine write tools. One of them,
enrich_article, refreshes a live article’s title, description and FAQs from that article’s own Search Console queries. - Max, $399 a month: higher data and MCP allowances, for agencies and many brands.
The 3-day free trial covers the app but not the MCP server; MCP access starts when your plan does. The pricing page lists every limit.
Get the server
Drafted with AI assistance from our own research and Search Console data, and reviewed by Rahul A before publishing. Rules and prices change; check the linked official source before you act.
