---
title: "Use Claude Code for Search Console analysis"
description: "Connect Claude Code to gscdump MCP, verify registered site access, and combine search data with a careful review of local content."
canonical_url: "https://gscdump.com/learn-google-search-console/ai-agents/claude-code-seo"
last_updated: "2026-09-11"
---

Claude Code is useful when a Search Console question leads back to your site's files. You can ask it to find pages losing clicks, locate the relevant content, and propose an edit you can review.

Start by verifying the data connection. A server being added to the configuration does not mean authorization or an analytics request succeeded.

## Add the hosted server

You need Claude Code installed and a gscdump account with your Google Search Console connection set up. Register the site through [Add a site](/app/sites/add) before asking the client to list it.

Run this in your terminal:

```bash
claude mcp add \
  --transport http --scope user \
  gscdump https://gscdump.com/mcp
```

User scope makes the server available across your local projects. Then inspect its status:

```bash
claude mcp get gscdump
```

Open Claude Code and run `/mcp`. Select gscdump and complete the browser authorization flow if prompted. These commands follow [Claude Code's MCP documentation](https://code.claude.com/docs/en/mcp).

If a server named `gscdump` already exists, inspect it before adding another. If it needs authentication, complete that step. If it cannot connect, resolve the connection error before asking for a report.

## Verify the site first

Use this prompt in Claude Code:

```text
Use gscdump's list-sites tool.
Show the registered sites and ask me which one to use.
For that site, show its siteId, sync status, and recorded date range.
Explain any missing or unavailable data before continuing.
Do not edit files.
```

Use the returned public `siteId` for the site you choose. Its scope can differ from its source Search Console property. Do not assume the tool result covers the whole property.

The recorded range gives you a starting point for choosing dates. It does not prove that every day is complete. The [MCP guide](/learn-google-search-console/ai-agents/mcp-server) explains the tool's coverage and error boundaries.

## Review one page with the repository open

After confirming comparable dates, use a bounded request. Replace the periods below with dates supported by your data:

```text
Compare page clicks for 1–7 September 2026 with 25–31 August 2026.
Keep the siteId and other report settings identical.
Use get-pages for each period and state the requested row limit.
Compare the union of returned pages. Mark absent rows as unknown,
not zero, unless another result establishes zero clicks.

Show the largest observed losses and the source values.
Ask me to choose one page. Then locate its source file in this repo.
Explain what the existing content covers and what needs checking.
Propose an edit, but do not change files yet.
```

This separates an observed change from a proposed explanation. A page can lose clicks because demand changed, because its search appearance changed, or for reasons the returned rows cannot identify.

Reading the source file adds context. It still doesn't prove that a missing paragraph caused a traffic decline. Ask Claude to label content recommendations as hypotheses until you have supporting evidence.

## Check the answer before accepting an edit

You should be able to trace each number to a tool result and each content claim to the file it read. Look for dates, filters, row limits, and missing-data notes in the output.

For a query appearing across several pages, ask which pages appeared and how their results changed. Multiple ranking pages alone don't prove harmful keyword cannibalization.

For an indexing question, ask Claude to inspect one exact URL with `inspect-url`. Google's [API reports the indexed version](https://developers.google.com/webmaster-tools/v1/urlInspection.index/inspect), so inspect the live page separately when the issue concerns a recent edit.

When you approve a proposed change, keep the normal repository review and testing steps. The [workflow prompts](/learn-google-search-console/ai-agents/seo-workflows) help you reuse the analysis without turning every traffic movement into an automatic content edit.

## Sitemap

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