"Which pages lost clicks after the March core update, and did their positions drop or just their CTR?" Once Search Console is wired up over MCP, that's a single question you type into Claude Code. Without it, it's twenty minutes of date pickers and CSV exports.
MCP (Model Context Protocol) is an open standard that lets tools like Claude Code, Claude Desktop, and Cursor call outside services mid-conversation. A server exposes typed tools, the model decides when to call them, and the results land back in context. For Search Console that means the model reads your actual click and impression data instead of guessing.
Two ways to wire it up
You can run an open-source GSC MCP server locally. There are at least eight on GitHub, most of them wrapping the Search Analytics API with a service account. They cost nothing and work fine for ad-hoc queries. They also inherit every limit of the live API, including the 16-month retention window and its row caps.
The other route is a hosted server backed by an archive, which is what Month17 is. Same protocol, same tools in your editor, but queries hit a permanent copy of your data instead of the live API. Year-over-year questions keep working after Google's window slides past.
The config
For Claude Code, one command:
claude mcp add --transport http month17 https://month17.com/mcp/ \
--header "Authorization: Bearer m17_your_api_key"For Claude Desktop or Cursor, the JSON equivalent:
{
"mcpServers": {
"month17": {
"url": "https://month17.com/mcp/",
"headers": {
"Authorization": "Bearer m17_your_api_key"
}
}
}
}
Restart the client and the tools show up automatically. Month17 exposes 22 of them, from raw search_analytics queries to prepared analyses like content_decay and cannibalization_check.
Prompts that pull their weight
- "Compare clicks and average position for /pricing/ this quarter against the same quarter last year."
- "Which queries have two of my pages competing against each other?"
- "List pages whose clicks declined for three consecutive months, worst first."
- "Show the impact of each confirmed Google core update on my traffic over the last two years."
Specific beats vague. Naming a URL, a metric, and a date range gets you an answer on the first call; "how is my SEO doing" gets you a fishing expedition across five tool calls.
Caveats worth knowing
GSC data runs two to three days behind, so "this week" is always partial. The API returns at most 50,000 rows per day per data type for a property, which trims the long tail on large sites. And a local server against the live API still has the archive problem: the window keeps sliding. MCP makes your search data easier to reach. It doesn't make it last longer. For that you need an actual archive.