Skip to content

MCP server (Node) · v1.0.0 · MIT

An SEO MCP server for Claude, Cursor and Claude Code

crawlcove-mcp is a Model Context Protocol server. Once it is connected, Claude Desktop, Claude Code, Cursor or any other MCP client can crawl a site live, or load a Crawl Cove desktop app export from disk, and answer questions from the crawl rather than guessing: which pages have no meta description, what links to the broken pages, which URLs are noindex. There is nothing to install; MCP clients run it with npx straight from GitHub.

crawlcove-mcp on GitHub Start your 14-day free trial

Free · MIT-licensed · README checked September 2026

Install and run 

# Claude Code:
claude mcp add crawlcove -- npx -y github:CrawlCove/crawlcove-mcp

# Claude Desktop (claude_desktop_config.json) or Cursor (.cursor/mcp.json):
{
  "mcpServers": {
    "crawlcove": {
      "command": "npx",
      "args": ["-y", "github:CrawlCove/crawlcove-mcp"]
    }
  }
}

Things to ask 

# things to ask once it is connected
"Crawl https://staging.example.com and list every issue."
"Which pages are missing meta descriptions? Draft one for each."
"What links to the 404s?"
"Load ~/Downloads/crawl-export.json and show me the noindex pages."

The README in the repo is the authoritative reference; the commands above are copied from it as of September 2026. If they disagree, the README is newer.

The tools it exposes

  • crawl_site: a live, same-origin, breadth-first crawl that respects robots.txt, up to 200 pages, which becomes the active dataset.
  • load_export: reads a Crawl Cove desktop app JSON export or a CLI result from disk; the 200-page limit of a live crawl does not apply to a loaded export.
  • get_issues: broken pages, missing or long or duplicate titles, missing or long meta descriptions, missing or multiple H1s, noindex, redirect chains and missing canonicals.
  • get_page: everything recorded about one URL, including redirect hops and the broken links out of it.
  • list_broken_links: broken internal links as source page to broken target, so you know which page to fix.
  • list_pages: pages by status, indexability or URL substring.

Limits worth knowing

  • Live crawls stop at 200 pages. For a whole site, run the crawl in the desktop app and load its export instead.
  • Requires Node 18 or later on the machine running the MCP client. The first start takes a few seconds while npx fetches the package.

Works with Crawl Cove

Crawl Cove MCP server covers the checks it lists above. The desktop app runs the full site audit: every page, every finding ranked by impact, history over time and Search Console data alongside. See the desktop checker or compare the plans.

Start your free trial
Related crawlcove-cli crawlcove-export-spec crawlcove-action All open-source repos

Frequently asked questions

Does the assistant send my crawl data anywhere?
The server runs on your machine and returns crawl results to the MCP client you connected it to. What that client then sends to its model is governed by the client and the model provider, exactly as it is for any other MCP tool you use.
Why is the live crawl capped at 200 pages?
So a question does not turn into a twenty-minute crawl inside a chat. For anything larger, crawl in the Crawl Cove desktop app and use load_export on the JSON it produces, which is not subject to the 200-page limit.
Which clients does it work with?
Any MCP client. The README carries the exact config for Claude Desktop, Claude Code and Cursor; the same npx command works anywhere an MCP server can be declared.

Stop guessing.
Start fixing.

Crawl Cove runs on your machine, connects to your real ranking data, and tells you exactly what to fix first. No per-feature paywalls, no spreadsheets, no guesswork.

Start your 14-day free trial
Compare all plans

28 days risk-free · No card required