MCP Scorecard

Mission StatementGitHub
← All posts

Screaming Frog Meets MCP: The SEO Spider Gets an AI Copilot

An MCP server wraps the Screaming Frog SEO Spider CLI — 8 tools for launching crawls, exporting data, and analyzing results through conversation. One catch: you have to close the GUI first.
io.github.bzsasson

Screaming Frog SEO Spider is the industry-standard website crawler for SEO professionals. If you've audited a website's technical SEO, you've probably used it. io.github.bzsasson/screaming-frog-mcp now lets you drive it through Claude.

How It Works

The MCP server wraps Screaming Frog's CLI executable, invoking it in headless mode. Eight tools cover the full crawl lifecycle:

  • sf_check — verify installation and license
  • crawl_site — launch headless background crawls
  • crawl_status — monitor progress in real time
  • list_crawls — show saved crawls with database IDs
  • export_crawl — generate CSV exports (all native export tabs, bulk exports, report formats)
  • read_crawl_data — query exported data with filtering and pagination
  • delete_crawl / storage_summary — housekeeping

The workflow this enables: "Crawl example.com, then show me all pages with 4xx status codes and missing meta descriptions" — spoken in natural language, executed through the same engine SEO professionals already trust.

The Single-Process Constraint

Screaming Frog uses an internal database that only supports single-process access. You must close the GUI before the MCP server can access crawl data. The recommended pattern: run complex crawls in the GUI with custom configurations, close the app, then use the MCP server for analysis and data extraction. This "GUI for configuration, AI for analysis" split is actually ergonomically sensible — it plays to each interface's strengths.

What to Know

Built by Boaz Sasson, who also maintains a separate MCP gateway project for optimizing token usage across multiple servers. Requires Screaming Frog v16+ and a paid license (free version caps at 500 URLs). Python, MIT-licensed, published on PyPI. Brand new — 1 star, version 0.1.0, published February 28, 2026.

Minimal dependencies: just the MCP SDK and python-dotenv. The server itself is a focused wrapper — the heavy lifting happens in Screaming Frog's engine, which has years of battle-testing behind it.

Score: 58. No flags. MIT.

Sources: Boaz Sasson — GitHub · screaming-frog-mcp — repo · Screaming Frog — website · Scorecard: io.github.bzsasson (score 58)

← Fully Homomorphic Encryption as an MCP Tool: Compute on Data You Can't SeeA Linter for Ideas: Does Your Project Already Exist? →