browser-automation

name	browser-automation
description	Automate web browsers using Playwright. Take screenshots, extract content from JavaScript-heavy pages, fill forms, click buttons, and run custom browser scripts. Use when the user needs to interact with web pages, test web apps, or scrape dynamic content.

Browser Automation

You can control a real web browser (Chrome on the host machine) using Playwright. xpressclaw launches Chrome with remote debugging, and Playwright inside your container connects to it via CDP.

Tools Available

• browser_screenshot(url, file_name, wait_for, full_page) — Take a screenshot
• browser_fetch(url, selector, wait_for) — Extract text from JS-rendered pages
• browser_run(script) — Run a custom Playwright Python script

Quick Tools

Screenshot

browser_screenshot(url="https://example.com", file_name="example.png", full_page=True)

Screenshots are saved to /workspace/screenshots/.

Fetch Page Content

browser_fetch(url="https://news.ycombinator.com", selector="tr.athing .titleline > a")

Extracts text from elements matching the CSS selector. Renders JavaScript first.

Custom Scripts

For complex automation, use browser_run with a Playwright script. The browser is already running — connect via CDP:

from playwright.sync_api import sync_playwright
import os

with sync_playwright() as p:
    browser = p.chromium.connect_over_cdp(os.environ["CDP_URL"])
    context = browser.new_context()
    page = context.new_page()

    # Navigate
    page.goto("https://example.com")

    # Extract text
    title = page.text_content("h1")
    print(f"Title: {title}")

    # Fill a form
    page.fill("input#email", "[e-mail]")
    page.click("button#submit")

    # Screenshot
    page.screenshot(path=os.environ["SCREENSHOTS_DIR"] + "/result.png")

    context.close()

Use $CDP_URL and $SCREENSHOTS_DIR environment variables in your scripts.

Rules

• Chrome runs on the host machine — Playwright in the container connects via CDP
• Use $CDP_URL to connect: p.chromium.connect_over_cdp(os.environ["CDP_URL"])
• Always create a new context and close it when done
• Screenshots saved to /workspace/screenshots/
• The browser is visible on the user's screen — they can see what's happening