The schema tool paired with extract is the workflow that sets this apart from a plain fetcher: describe the fields you want, get a JSON Schema, and pull data that already matches it. The monitor tools push it further — instead of re-scraping on your own cron, monitor_create schedules the check and monitor_activity tells you which runs actually saw the page change.
ScrapeGraphAI's official server against their API v2: 17 tools for AI-powered extraction, search, crawling and scheduled monitoring, plus schema generation and account reads. Runs from PyPI over stdio.
- extract pulls structured data from a webpage with AI, and scrape fetches raw page content when you want the whole thing
- search runs an AI web search that returns structured data rather than a list of links to chase
- A full async crawl lifecycle — crawl_start kicks off a multi-page SmartCrawler and returns an id, crawl_get_status polls it until completed, and crawl_stop and crawl_resume pause and continue a running job
- schema generates or augments a JSON Schema from a natural-language prompt, so a structured extraction has a shape to fill before you run it
- Scheduled monitoring as its own surface: monitor_create makes a job (wrapping your prompt and optional output_schema into the v2 formats array), and monitor_list, monitor_get, monitor_pause, monitor_resume and monitor_delete manage them
- monitor_activity polls a monitor's per-run tick history — each run's status, whether anything changed, elapsed time and diffs — with a cursor when there is more
- credits returns your remaining API credits, and history lists recent requests by page, limit and service
A ScrapeGraph AI API key, held in SGAI_API_KEY (or passed through the MCP config). Python, run from the scrapegraph-mcp PyPI package over stdio. It targets API v2 at https://v2-api.scrapegraphai.com/api by default; SGAI_API_URL overrides the base URL and SGAI_TIMEOUT sets the request timeout, which defaults to 120 seconds.
One command plus a key — npx -y @smithery/cli install @ScrapeGraphAI/scrapegraph-mcp --client claude, then supply credentials