Playwright Browser Automation

The AI agent can use Playwright to interact with web applications through a real headless Chromium browser. This is essential for testing modern JavaScript-heavy applications, SPAs, and any page where curl only returns empty loader HTML.

What is Playwright?

Playwright is a browser automation framework that controls headless Chromium inside the Kali sandbox container. It is exposed to the AI agent as the execute_playwright MCP tool on port 8005.

Two Modes

Content Extraction Mode

The simplest way to use Playwright. Provide a URL and optionally a CSS selector and format.

How to ask the agent:

• "Use Playwright to get the content of http://10.0.0.5:3000"
• "Extract the login form HTML from http://10.0.0.5/login"
• "Get the text inside the .dashboard-panel element"

Parameters:

The browser navigates to the URL, waits for the page to finish loading (network idle), and extracts the content.

Script Mode

For complex multi-step interactions. You provide Python code that uses Playwright's sync API. Three variables are pre-initialized for you:

Use print() to send output back to the agent.

How to ask the agent:

• "Run a Playwright script to log in with admin/admin and show the dashboard"
• "Use Playwright to fill the search form with an XSS payload and show the response"
• "Write a Playwright script to extract all links from the target homepage"

Example Chat Prompts

Here are some things you can ask the agent in the chat:

Content Extraction

"Use Playwright to extract the page content from http://10.0.0.5:8080"

"Get the HTML of the login form at http://10.0.0.5/login using Playwright"

"Extract the navigation menu from the target homepage"

Login Testing

"Use Playwright to try logging in with admin/password on http://10.0.0.5/login"

"Run a Playwright script to fill the login form and check if we get access"

XSS Testing

"Test XSS by injecting a script tag into the search field using Playwright"

"Use Playwright to submit <img src=x onerror=alert(1)> into the comment form"

Page Analysis

"Use Playwright to check the response headers of http://10.0.0.5 for security headers"

"Extract all JavaScript files referenced on the target page"

Confirmation Required

Playwright is classified as a dangerous tool. Every time the agent wants to use it, you will see a confirmation prompt in the chat. You must approve the execution before it runs.

This is because Playwright can interact with web applications -- filling forms, clicking buttons, submitting data -- which could have side effects on the target.

Stealth Mode

When stealth mode is enabled, Playwright usage is restricted:

• Single URL per call -- no automated crawling or bulk scraping
• Max 2 form submissions per target -- then the agent stops and asks you
• No credential spraying -- single login attempts only
• No fuzzing via browser -- use dedicated tools instead

Limitations

Troubleshooting

"Script timed out after 45 seconds"

The page is taking too long to reach network idle. Try:

• Targeting a specific element with a CSS selector instead of the full page
• Using script mode with a shorter wait_for_load_state('domcontentloaded') instead of networkidle

"No element found matching selector"

The CSS selector does not match anything on the page. The page may:

• Require authentication first (use script mode to login)
• Load content dynamically after initial render (add page.wait_for_selector('selector'))
• Use shadow DOM (access via page.locator('selector'))

Empty output

• The page may be a SPA that needs more time to render -- use script mode and add explicit waits
• Content may be inside an iframe -- use page.frame() to access it

Technical Reference

For full technical documentation including architecture diagrams, Docker setup, and implementation details, see README.PLAYWRIGHT.md.