Repository navigation
First public release.
An MCP server that gives Claude web scraping powers. Free, local, no API key, no cloud account.
24 tools covering pages, whole-site crawls, products, contacts, Shopify catalogues, PDFs, places, page interaction and change tracking.
Four escalating fetch tiers — plain httpx, TLS fingerprint impersonation via curl_cffi, Playwright, then stealth Playwright. It only escalates when it detects an actual refusal, so the common case stays cheap. Verified live against ASOS, eBay, Amazon, Booking.com, Trustpilot, Rightmove, Indeed, Zoopla, LinkedIn, Reddit and X.
Block detection that reads the body, not the status code. Every failure found in live testing was an HTTP 200 that looked like success, and one site served a real page under a 403.
Context economy. Articles are boilerplate-stripped (a BBC news page goes 20,513 to 3,198 chars) and large results land in a SQLite table you query with read-only SQL.
311 unit tests plus 9 integration tests, green on Python 3.10 through 3.13. MIT.
pipx install git+https://github.com/maccydee/cute-web-scraperKnown limitation: Google Maps is a consent wall rather than bot detection, so no tier reaches it. Place lookups use OpenStreetMap instead.