examples

Examples

Every example is runnable. They all need an API key — python -m antibrow login once, or export ANTIBROW_API_KEY=... — and the first run downloads the browser kernel (~190 MB, cached afterwards).

File	What it shows	Extra install
01_basic.py	Launch, browse, read the fingerprint back, screenshot CreepJS	—
02_persistent_profile.py	Profiles: frozen identity + persistent storage. Run it twice	—
03_proxy_geoip.py	HTTP/SOCKS5 proxies, and making timezone + WebRTC follow the exit IP	—
04_playwright.py	Locators, routing, init scripts, cookies — plain Playwright	—
05_puppeteer_style.py	Puppeteer/Node idiom → Python mapping, and attaching Node over CDP	—
06_browser_use.py	An LLM agent driving the browser	pip install browser-use
07_crawl4ai.py	crawl4ai's markdown pipeline over a fingerprinted browser	pip install crawl4ai
08_scrapling.py	Scrapling's adaptive selectors over the same session	pip install scrapling
09_mcp_server.py	A full MCP server: any agent gets a stealth browser	pip install "antibrow[mcp]"
10_docker/	Container image, headful under Xvfb, cached kernel volume	Docker

python examples/01_basic.py
ANTIBROW_PROXY="socks5://user:pass@host:1080" python examples/03_proxy_geoip.py

Notes

• Linux/Docker: run against an Xvfb display (see the Dockerfile; xvfb-run hangs under a container's PID 1). headless=True is a Windows-only trick (off-screen window); on Linux a virtual display is the honest equivalent, and real --headless is avoided because it is itself a fingerprint.
• Third-party APIs move. 06–08 target the integrations' current APIs (browser-use, crawl4ai, Scrapling). If one changes, the stable part is always session.cdp_url / session.cdp_endpoint — hand that to whatever the library now calls its "connect to an existing browser" option.
• Concurrency: each running browser consumes one slot from your plan's cap. Free is 1; the kernel enforces it and raises ConcurrencyLimitError.