---
title: "Google Search Console MCP server · gscdump"
canonical_url: "https://gscdump.com/mcp"
last_updated: "2026-10-08T14:05:13.533Z"
meta:
  description: "gscdump is a hosted Google Search Console MCP server. Connect ChatGPT or Claude with OAuth and query history kept past Google's 16-month window."
  "og:description": "gscdump is a hosted Google Search Console MCP server. Connect ChatGPT or Claude with OAuth and query history kept past Google's 16-month window."
  "og:title": "Google Search Console MCP server · gscdump"
---

# Google Search Console MCP server

gscdump is a hosted MCP server over your synced Search Console record, kept past Google's 16-month window. Connect with OAuth, then ask ChatGPT or Claude about the whole record.

[**Join the waitlist **](https://gscdump.com/#get-access) [**Check your connection **](https://gscdump.com/learn-google-search-console/ai-agents/mcp-server)

Plugins → Add → Create MCP App

```
Name
gscdump

Server URL
https://gscdump.com/mcp

Authentication
OAuth
```

Needs developer mode on the web. Tick the risk checkbox, then Create.

## What your agent can ask

16 tools

The seven analytics reads answer from your preserved record, sitemap listings come from the synced sitemap tables, indexing reads come from stored URL Inspection verdicts, and URL inspection calls Google at request time. Each tool publishes its input schema, so the client picks the matching call from a plain question.

<dl>

<dt>list-sites</dt>
<dd>Your connected sites, each with sync status and the oldest and newest dates synced.</dd>

<dt>list-sitemaps</dt>
<dd>Sitemaps for a site: URL counts, errors, warnings, and last-download timestamps.</dd>

<dt>get-sitemap-submission</dt>
<dd>The sitemap gscdump would submit to Google for a site, and what blocks it: write access, Owner or Full permission, or no sitemap found.</dd>

<dt>list-bing-sites</dt>
<dd>Bing state per site: collecting, DNS record needed, ready to link, or Bing authorization needed. Bing is in private preview.</dd>

<dt>get-analytics</dt>
<dd>Totals for clicks, impressions, CTR, and position over a date range, with an optional daily series and an optional trajectory: the shape of the whole record, such as a launch spike followed by a cliff.</dd>

<dt>get-keywords</dt>
<dd>Aggregated performance per query.</dd>

<dt>get-keyword-details</dt>
<dd>One Search Console query in depth: totals, daily series, and pages that appeared for it.</dd>

<dt>get-pages</dt>
<dd>Aggregated performance per URL.</dd>

<dt>get-page-details</dt>
<dd>One page in depth: totals, daily series, and Search Console queries that showed it.</dd>

<dt>get-countries</dt>
<dd>Performance broken down by country.</dd>

<dt>get-devices</dt>
<dd>Performance broken down by device: mobile, desktop, tablet.</dd>

<dt>inspect-url</dt>
<dd>Google's indexed snapshot of a URL: verdict, coverage state, canonical signals, last crawl time. Counts against Google's 2,000 inspections per site per day.</dd>

<dt>get-indexing</dt>
<dd>Stored URL Inspection verdicts per day, counted by coverage state: unknown to Google, discovered, crawled, and indexed. Says when gscdump counted them and how old they are.</dd>

<dt>list-watched-urls</dt>
<dd>Watched URLs and their Checkpoints: the coverage state each weekly inspection found.</dd>

<dt>run-report</dt>
<dd>Composed analysis reports that fan out across multiple analyzers. Cached by input hash, regenerated on demand.</dd>

<dt>gsc-discover-hourly</dt>
<dd>Hourly Discover clicks, impressions, and position per URL for one calendar day, on sites with hourly Discover sync enabled.</dd></dl>

## Hosted and local

Hosted

### **The platform **

gscdump.com runs the sync and holds the record in open formats. OAuth links each MCP client to your gscdump account. Legacy clients can still send an API key.

https://gscdump.com/mcp

$ Local

### **The local twin **

The CLI writes exports to your disk. In Local mode, its stdio MCP server calls Google with your own credentials. The same CLI runs in Hosted mode: `--mode hosted` reads your platform record and never calls Google.

```
npx @gscdump/cli init npx @gscdump/cli mcp --mode local
```

ChatGPT, Claude, and Claude Code can use OAuth. Cursor and Codex can use the API key fallback. The local server uses stdio.

## Compared with self-hosted Search Console MCP servers

Most Search Console MCP servers run on your machine and pass each question to the Search Console API. That works for recent data. The differences show up in what the answer can reach.

<dl>

<dt>History</dt>
<dd>A live API proxy reaches only the 16 months Google keeps. gscdump answers from its synced record, which keeps the days it has synced after Google deletes them.</dd>

<dt>Setup</dt>
<dd>A self-hosted server usually needs its own Google Cloud credentials and a process on each machine. gscdump is one URL with OAuth, and the same record reaches every client.</dd>

<dt>Scope</dt>
<dd>Every tool call re-checks your access per user and per Site. A Site covers the exact host you registered, so totals can sit below a broader Search Console property.</dd></dl>

Prefer to run it yourself? The CLI above has a local MCP server that queries live Google data with your own authentication.

## What the agent can trust

An agent should be able to tell, from the response alone, whether to trust it. The server states what it knows:

<dl>

<dt>meta.dataRange</dt>
<dd>Every analytics answer carries the site's sync status and the synced date window the numbers cover.</dd>

<dt>gscdump.com/retry</dt>
<dd>Every failed call states whether an identical retry can succeed, as a structured hint in the result's <code>_meta</code> and in prose. Transient backpressure comes with a wait in seconds; a broken query says so plainly instead of inviting a retry loop.</dd>

<dt>lastDownloaded</dt>
<dd>Sitemap rows carry their download timestamps, error and warning counts, and the last error, so stale reads as stale.</dd>

<dt>lastCrawlTime</dt>
<dd>URL inspection returns Google's indexed snapshot, dated by when Google last crawled the page rather than presented as live.</dd></dl>

Capture time ships today on the sitemap and inspection surfaces above; analytics answers carry their synced window instead. A Site covers the exact registered host, so its analytics can exclude traffic shown by a broader Search Console property. Apply a matching host filter when comparing totals. Source limits and sync coverage can also cause differences.

## Questions

The platform is free during beta, MCP included. The CLI and its local server are MIT licensed on npm and stay free.

Only sites your Google account can access and that you have connected to gscdump. Every tool call re-checks access per user and per site. You can revoke Google access or disconnect a site at any time.

For platform MCP, add https://gscdump.com/mcp to your client and sign in to read synced history. CLI MCP runs locally and queries live Google data with your selected authentication. CLI exports stay on your disk for SQL queries.

ChatGPT and Claude use OAuth, so they do not need an API key. Cursor and Codex can use a key from settings.

Private beta. Join the waitlist.

Signups open gradually during beta. Leave your email on the home page and an invite follows when a slot opens.

[**Join the waitlist **](https://gscdump.com/#get-access) [**Install the CLI **](https://www.npmjs.com/package/@gscdump/cli)

## Sitemap

See the full [sitemap](/sitemap.md) for all pages.
