How to Install search-console-mcp: Ask AI Why Your Traffic Dropped (Free)
Tools
Step-by-step
Search-console-mcp
Trigger Marketing
The Trigger team
⊹
Your traffic dipped. So you open Search Console, export a CSV, open a spreadsheet, squint at it, and lose your evening.
Or you type one sentence into your AI assistant: "My traffic dropped this week. Which pages did it?" And it answers with your real numbers.
That second option is free. It's called search-console-mcp, and here's how to set it up.
Quick words (one line each):
Google Search Console: Google's free dashboard that shows which searches bring people to your site.
MCP (Model Context Protocol): a plug that lets an AI app (Claude, Cursor and others) talk to another tool.
MCP server: the small program on the other end of that plug. This tool is one.
Terminal: the text window where you type commands instead of clicking.
npx: a command that downloads and runs a Node.js tool in one go.
Why bother? Because AI guesses and data doesn't
John Morabito, who runs SEO agencies, said it plainly on the Search With Candour podcast (Oct 26, 2026): plug real data tools into your AI instead of trusting its own crawling or keyword volumes (0:46:40-0:48:30).
search-console-mcp is exactly that kind of plug. It reads your real Search Console data, does the math (drops, cannibalization, quick wins) itself, and then hands the answer to your AI.
Step 1: Do a 60-second safety check
Never install a GitHub tool blind. Here's what search-console-mcp showed when we checked it on Oct 28, 2026:
Licence: MIT. Free to use, including for client work.
Last update: Oct 21, 2026 (version 2.1.4). Very active.
Stars: 312 (as of Oct 28, 2026).
Read-only by default. The README says tools that change things (submitting or deleting sitemaps, managing sites) need an extra write permission that you choose to give.
Where your login lives: the README says tokens go into your computer's own keychain (Windows Credential Manager, macOS Keychain or Linux Secret Service).
Our test: we installed it in a scratch folder on a test computer. It listed 25 tools and correctly said "Google not connected" before any login.
Step 2: Run the setup (one command)
You need Node.js installed. In a terminal:
npx search-console-mcp setup
A browser window opens. Sign in with the Google account that owns your Search Console property and approve access. Done.
Step 3: Tell your AI app about it
Open your AI app's MCP settings file (Claude Desktop, Cursor and others have one) and add this, exactly as the README shows it:
{
"mcpServers": {
"search-console": {
"command": "npx",
"args": ["search-console-mcp"]
}
}
}
Restart the app. That's the whole install.
Claude Desktop shortcut: the README also offers a one-click .mcpb bundle on the repo's Releases page. Download it and drag it in.
Step 4: Ask it real questions
The README gives these starter prompts. Paste them straight in:
"My traffic dropped this week vs. last. Find exactly when it started and which pages are responsible."
"Find keywords for example.com ranking positions 8-15 with 1,000+ impressions - my best quick wins."
"Check for keyword cannibalization - are two of my pages competing for the same query?"
"Run seo_audit on my top pages: which have high search visibility but poor CTR?"
"Run a full SEO health check (site_health_check), segmented by Brand vs Non-Brand."
Cannibalization, by the way, means two of your own pages fighting over the same search. Nobody wins that fight.
Step 5: Turn the answers into work
Quick wins list (positions 8-15): these are your best pages to refresh. Brian Dean on the Semrush channel (Oct 22, 2026) quoted an industry study: pages refreshed in the last 3 months were 3x more likely to be cited by AI (0:02:40-0:04:00).
Brand vs non-brand split: Exposure Ninja (Sep 28, 2026) quoted a Similarweb study: after an AI recommends a brand, 55% of the time the person then Googles the brand (0:04:00-0:04:40). So a rise in branded searches is where your AI wins show up. Watch it.
Step 6: Add Bing and GA4 (optional)
The README lists these extra hookups:
Bing Webmaster Tools: set a Bing API key (BING_API_KEY).
Google Analytics 4: run npx search-console-mcp setup --engine=ga4 (uses a Google service account).
AdSense: only if your site runs ads.
Step 7: Use it without an AI app (bonus)
The tools also run straight from the terminal. The README's example:
npx search-console-mcp run seo_audit --siteUrl=https://example.com --type=quick_wins
Add --format=csv or --format=table to some commands for a spreadsheet-ready answer.
Step 8: Know what's real and what's a guess
genai_query_insights flags long, chat-style searches that look like AI traffic. The README says straight up that Google and Bing give no official AI-citation data, so this is a smart guess from normal query data. Treat it as a hint, not a report.
Keep it read-only unless you truly need the write tools. Fewer permissions, fewer surprises.
What it costs
Nothing. MIT licence, and the Search Console, GA4 and Bing Webmaster APIs are free.
Do this today
Install Node.js if you don't have it.
Run npx search-console-mcp setup and sign in with the Google account that owns your site.
Paste the MCP config from Step 3 into your AI app and restart it.
Ask: "Find my keywords ranking positions 8-15 with 1,000+ impressions."
Pick the top 3 pages from that list and refresh one real section on each.
Related reading
Sources
Don't worry if this seems like it's a pain in the ass for someone who doesn't build websites. That's because it is. The good news: just click the link below, pull the trigger, and we'll take care of the rest.
Share This Article
