scrapling - SKILL.md Agent Skill

name: scrapling

description: "Web scraping using Scrapling — a Python framework with anti-bot bypass (Cloudflare Turnstile, fingerprint spoofing), adaptive element tracking, stealth headless browser, and full CSS/XPath extraction. Use when web_fetch fails (Cloudflare, JS-rendered pages), or when extracting structured data from websites (prices, articles, lists). Supports HTTP, stealth, and full browser modes. Source: github.com/D4Vinci/Scrapling (PyPI: scrapling). Only use on sites you have permission to scrape."

license: MIT

metadata:

source: https://github.com/D4Vinci/Scrapling

pypi: https://pypi.org/project/scrapling/

Scrapling Skill

Source: https://github.com/D4Vinci/Scrapling (open source, MIT-like license)

PyPI: scrapling — install before first use (see below)

⚠️ Only scrape sites you have permission to access. Respect robots.txt and Terms of Service. Do not use stealth modes to bypass paywalls or access restricted content without authorization.

Installation (one-time, confirm with user before running)


pip install scrapling[all]

patchright install chromium  # required for stealth/dynamic modes

scrapling[all] installs patchright (a stealth fork of Playwright, bundled as a PyPI package — not a typo), curl_cffi, MCP server deps, and IPython shell.
patchright install chromium downloads Chromium (~100 MB) via patchright's own installer (same mechanism as playwright install chromium).
Confirm with user before running — installs ~200 MB of dependencies and browser binaries.

Script

scripts/scrape.py — CLI wrapper for all three fetcher modes.


# Basic fetch (text output)

python3 ~/skills/scrapling/scripts/scrape.py <url> -q



# CSS selector extraction

python3 ~/skills/scrapling/scripts/scrape.py <url> --selector ".class" -q



# Stealth mode (Cloudflare bypass) — only on sites you're authorized to access

python3 ~/skills/scrapling/scripts/scrape.py <url> --mode stealth -q



# JSON output

python3 ~/skills/scrapling/scripts/scrape.py <url> --selector "h2" --json -q

Fetcher Modes

http (default) — Fast HTTP with browser TLS fingerprint spoofing. Most sites.
stealth — Headless Chrome with anti-detect. For Cloudflare/anti-bot.
dynamic — Full Playwright browser. For heavy JS SPAs.

When to Use Each Mode

web_fetch returns 403/429/Cloudflare challenge → use --mode stealth
Page content requires JS execution → use --mode dynamic
Regular site, just need text/data → use --mode http (default)

Python Inline Usage

For custom logic beyond the CLI, write inline Python. See references/patterns.md for:

Adaptive scraping (auto_save / adaptive — saves element fingerprints locally)
Session/cookie handling
Async usage
XPath, find_similar, attribute extraction

Notes

MCP server (scrapling mcp): starts a local network service for AI-native scraping. Only start if explicitly needed and trusted — it exposes a local HTTP server.
auto_save=True: persists element fingerprints to disk for adaptive re-scraping. Creates local state in working directory.
Stealth/dynamic modes use Chromium headless — no xvfb-run needed.
For large-scale crawls, use the Spider API (see Scrapling docs).