Playwright Browser Automation

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.

Feature	Description
JS rendering	Fully renders JavaScript -- handles React, Angular, Vue, and other SPAs
Form interaction	Fill inputs, click buttons, submit forms, handle CSRF tokens automatically
Content extraction	Extract visible text or raw HTML from any page or element
Custom scripting	Run multi-step browser workflows (login, navigate, extract, verify)
Chrome spoofing	Uses a Chrome 120 user-agent for realistic requests

---

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:

Parameter	Required	Default	Description
url	Yes	--	URL to navigate to
selector	No	entire page	CSS selector to target a specific element
format	No	text	text for visible text, html for raw HTML

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:

Variable	What it is
browser	Headless Chromium instance
context	Browser context with Chrome user-agent
page	A new page ready for navigation

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

Limitation	Details
Output size	Capped at 15,000 characters per extraction
Content mode timeout	45 seconds (page must finish loading)
Script mode timeout	60 seconds (total script execution)
No file downloads	Browser runs headless inside Docker -- no download handling
No screenshots	Output is text/HTML only, no image capture

---

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.