Better Search Console

Why This Exists

Every SEO tool that connects to an LLM has the same problem: the Google Search Console API returns 1,000 rows per request, your context window fills up fast, and you end up asking the model to reason about a tiny fraction of your data. The answers sound plausible but they're built on incomplete information.

This MCP server takes a different approach. It downloads your entire Search Console dataset into a local SQLite database, then gives Claude pre-built SQL queries for every standard SEO analysis type. Claude never sees the raw data. It sees query results: the top 50 declining pages, the queries ranking 5-20 with high impressions, the device breakdown for a specific URL pattern. Precise answers from complete data, using a few hundred tokens instead of tens of thousands.

What you get

One database per GSC property, synced in the background with full pagination (no row limits). Sixteen pre-built insight queries covering the things SEOs check repeatedly. Custom SQL for everything else. Automatic data retention that prunes low-value rows after each sync to keep databases from growing without bound. The data stays on your machine and the queries run in milliseconds.

Quick Start

Step 1: Set Up Google Credentials

You need a Google Cloud service account with Search Console API access. Here is how to create one from scratch.

Create a Google Cloud project and enable the API

1. Go to the Google Cloud Console

2. Create a new project (or select an existing one)

3. Go to APIs and Services > Library

4. Search for Google Search Console API and click Enable

Create a service account

1. Go to APIs and Services > Credentials

2. Click Create Credentials > Service account

3. Give it a name (e.g. search-console-mcp) and click Create and Continue

4. Skip the optional role and user access steps, click Done

5. Click on the service account you just created

6. Go to the Keys tab

7. Click Add Key > Create new key > JSON

8. Save the downloaded JSON file somewhere safe (e.g. ~/credentials/gsc-service-account.json)

For full details on creating credentials, see the Google Workspace credentials guide.

Grant the service account access to Search Console

1. Open the JSON key file and copy the client_email value (it looks like name@project.iam.gserviceaccount.com)

2. Go to Google Search Console

3. Select a property

4. Go to Settings > Users and permissions > Add user

5. Paste the service account email and set permission to Full

6. Repeat for each property you want to access

Step 2: Add to Claude Desktop

Add this to your Claude Desktop config file:

• Windows: %APPDATA%\Claude\claude_desktop_config.json

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "better-search-console": {
      "command": "npx",
      "args": ["-y", "@houtini/better-search-console"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/your-service-account.json"
      }
    }
  }
}

Replace /path/to/your-service-account.json with the actual path to the JSON key file you downloaded in Step 1.

Step 3: Restart Claude Desktop and go

Tell Claude: "Show me my search console data"

The setup tool handles the rest: discovers your properties, syncs them all in the background, and shows an overview grid with sparkline trends. Depending on how much data you have, initial sync takes anywhere from 30 seconds to a few minutes.

Environment Variables

Tools

Setup and Sync

Analysis

Insight Types

The get_insights tool supports these report types:

Data Retention

Large GSC properties generate millions of rows. A site with 50,000 pages indexed across 200+ countries accumulates data fast, and most of it is noise: zero-click impressions from countries you don't target, queries that appeared once and never again.

The retention system prunes this automatically after each sync. Recent data (last 90 days by default) is never touched. Older data is filtered by two rules:

Target countries (US, UK, EU, Australia, Canada): rows with zero clicks are kept only if they have 5+ impressions. This preserves queries with visibility potential whilst removing the long tail of single-impression noise.

Non-target countries: rows with zero clicks are deleted entirely. If nobody in Uzbekistan has ever clicked through on one of your results, that row isn't informing any decision.

Rows with clicks are never deleted regardless of age or country.

After deletion, the database is vacuumed to reclaim disk space and ANALYZE is run to update the query planner.

Manual Pruning

You can trigger this yourself with the prune_database tool. Preview mode shows what would be deleted without touching anything:

prune_database siteUrl="sc-domain:example.com" preview=true

Then run for real:

prune_database siteUrl="sc-domain:example.com"

Both recentDays (default: 90) and targetMinImpressions (default: 5) are configurable per run.

Results

On a site with 9.8 million rows and a 6GB database, the initial prune deleted 6.3 million rows (64%), reduced the database to 2.2GB, and preserved every single click. All impressions data for actionable queries remained intact. Subsequent syncs prune incrementally, so the database never bloats again.

Custom SQL

The query_gsc_data tool accepts any SELECT query against the search_analytics table. INSERT, UPDATE, DELETE, DROP, ALTER, CREATE, and PRAGMA are blocked. Queries without an explicit LIMIT are automatically capped at 10,000 rows.

Schema

search_analytics (
  date       TEXT,     -- YYYY-MM-DD
  query      TEXT,
  page       TEXT,
  device     TEXT,     -- DESKTOP, MOBILE, TABLET
  country    TEXT,     -- ISO 3166-1 alpha-3 lowercase
  clicks     INTEGER,
  impressions INTEGER,
  ctr        REAL,
  position   REAL
)

Example Queries

Find cannibalisation (multiple pages ranking for the same query):

SELECT query, COUNT(DISTINCT page) as pages, SUM(clicks) as clicks
FROM search_analytics
WHERE date >= '2025-01-01'
GROUP BY query
HAVING pages > 1
ORDER BY clicks DESC
LIMIT 20

Content decay detection:

SELECT page,
  SUM(CASE WHEN date >= date('now', '-28 days') THEN clicks END) as recent,
  SUM(CASE WHEN date BETWEEN date('now', '-56 days') AND date('now', '-29 days') THEN clicks END) as prior
FROM search_analytics
GROUP BY page
HAVING prior > 10
ORDER BY (recent * 1.0 / NULLIF(prior, 0)) ASC
LIMIT 20

Mobile vs desktop performance gap:

SELECT page, device,
  SUM(clicks) as clicks,
  AVG(position) as avg_position
FROM search_analytics
WHERE date >= date('now', '-28 days')
GROUP BY page, device
HAVING clicks > 5
ORDER BY page, device

Development

Building from Source

git clone https://github.com/houtini-ai/better-search-console.git
cd better-search-console
npm install
npm run build

Then configure Claude Desktop to use your local build:

{
  "mcpServers": {
    "better-search-console": {
      "command": "node",
      "args": ["/path/to/better-search-console/dist/index.js"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/service-account.json",
        "BSC_DATA_DIR": "/path/to/data-directory"
      }
    }
  }
}

Scripts

Troubleshooting

"No Google Search Console properties found"

The service account needs to be added as a user on each property. Go to Search Console > Settings > Users and permissions > Add user. Use the service account email from your JSON key file (it looks like name@project.iam.gserviceaccount.com). Grant "Full" permission level.

Sync takes a long time

Large properties with years of data and many countries can have millions of rows. Initial sync for a property with 10 million rows takes 5-10 minutes. Subsequent syncs are incremental (only fetching new dates) and complete in seconds.

Database files are large

Run prune_database with preview=true to see how much can be cleaned up. The default retention policy typically removes 50-70% of rows from properties with international traffic. The auto-prune after each sync prevents future bloat.

ext-apps UI not rendering

The interactive dashboards require Claude Desktop with ext-apps support. If you're using an older version or a different MCP client, the text-based fallback responses still contain all the data.

Licence

Apache-2.0. See LICENSE for details.

Disclaimer

This software is provided as-is. Your GSC data is stored locally in SQLite databases on your machine. The server communicates with Google's Search Console API using your service account credentials. Review Google's API terms of service and ensure your usage complies.

API keys and service account credentials are sensitive. Keep them out of version control, rotate them if exposed, and restrict service account permissions to the Search Console API only.

Built by Houtini for the Model Context Protocol community.