Google Search Console MCP server

The Google Search Console MCP server is a ready-made scenario. It sits between your Search Console account and an MCP client, so you can ask questions like "what dropped last month?" and get live data back.

You import the template, connect Google, copy the server URL into Claude, Cursor, or another client, and turn the scenario on. The server is read-only: it cannot change, add, or delete anything in Search Console.

What you get

Nine tools. You do not call them by name: the assistant picks one from your question.

Four of these go beyond the Search Console UI: period comparison by query, striking-distance keywords, low-CTR pages, and cannibalization.

Set up the server

Connect your client

Replace SERVER_URL and API_KEY with your values. If a client asks for a header, use Authorization: Bearer API_KEY.

{
  "mcpServers": {
    "gsc": {
      "type": "http",
      "url": "SERVER_URL",
      "headers": { "Authorization": "Bearer API_KEY" }
    }
  }
}

{
  "mcpServers": {
    "gsc": {
      "url": "SERVER_URL",
      "headers": { "Authorization": "Bearer API_KEY" }
    }
  }
}

[mcp_servers.gsc]
url = "SERVER_URL"
[mcp_servers.gsc.http_headers]
authorization = "Bearer API_KEY"

Setup for other MCP clients is the same idea: paste the Server URL and the API key. See Connecting to MCP Tools.

Connect Google

In Search Console, open Settings → Users and permissions and confirm your account is listed.

Full user is the practical minimum. With Restricted, inspect_url and sitemap_audit return permission errors.

Then open any Google Search Console step in the scenario, click Connect in the Connection field, choose the account, and approve. Do this once: every step shares the connection.

To check that it works, ask: List my GSC properties.

How to ask

• Name the site exactly. sc-domain:example.com for a domain property, or https://example.com/ with the trailing slash for a URL-prefix property. Copy it from list_properties.
• Give a date range. For example: the last 28 days, or 1 to 30 June. Without one, the assistant picks dates itself.
• Do not ask about yesterday. Search Console data lags two to four days. End the range three days back.

Starter prompts:

• Show the top 20 queries for sc-domain:example.com from 1 to 28 July
• Compare the last 28 days with the previous 28 for sc-domain:example.com
• Find striking-distance keywords for sc-domain:example.com over the last 28 days
• Which pages on sc-domain:example.com have below-average CTR in July?
• Check sc-domain:example.com for keyword cannibalization over the last 90 days
• Which queries drive traffic to https://example.com/pricing/ in July?
• Check the index status of https://example.com/pricing/
• Are there any errors in the sitemaps for sc-domain:example.com?

If the assistant picks the wrong tool, name it: use find_striking_distance for sc-domain:example.com.

How to read the tools

compare_periods. Name one period; the tool uses the same number of days immediately before. A negative position change means the rank improved (the number got smaller). Queries labelled new or lost may have crossed Google's privacy threshold rather than truly appeared or vanished.

get_search_by_page_query. Use the full URL as Search Console stores it, including the trailing slash. Asking about /pricing when the property stores https://example.com/pricing/ returns nothing.

find_striking_distance. Queries in positions 8 to 20 that already get impressions but few clicks. The potential-clicks figure is a priority signal, not a forecast.

find_low_ctr_pages. Pages whose CTR is below what is typical for their average position. Average position is averaged across every query the page ranks for. A weak CTR can also come from an AI Overview above the result, so treat the list as candidates, not a verdict.

keyword_cannibalization. Checks your 50 highest-volume queries (up to 25 competing pages each) to stay inside Google's daily quota. The conflict flag is true when the best-ranking page is not the one earning the clicks. Several pages ranking for one query is often normal.

inspect_url. Google allows 2,000 inspections per day per property.

sitemap_audit. Pay attention to any sitemap Google has never downloaded: usually unreachable or submitted with the wrong address.

Google limits

These limits come from Search Console, not from the server:

• Data lags two to four days; there is no live view
• Some rare queries are hidden for privacy, so query rows never add up to the totals in the Search Console UI
• Average position is an average across all impressions in the period, not where you rank right now
• History stops at 16 months
• A single response holds 25,000 rows at most, ordered by clicks; the long tail may be missing

Troubleshooting

Related pages

• MCP Nodes: how MCP Trigger and tools work
• Connecting to MCP Tools: connect other MCP clients