Browser Automation

Stealth browser automation with VNC viewing — navigate websites, fill forms, extract content, and scrape data via Chrome DevTools Protocol.

What You Get

• 18 MCP tools — navigation, form filling, screenshots, content extraction, scraping
• Docker container — headless Chrome with Xvfb + VNC viewer
• Stealth mode — human-like typing, click randomization, fingerprint spoofing
• Dashboard panel — live VNC view, container controls, saved sessions, installed skills
• Content extraction — article text, HTML tables, structured data, pagination
• FFFF filing skill — IRS Free File Fillable Forms automation (bundled)

Installation

1. Open the Extensions page in your Crow's Nest dashboard
2. Find Browser Automation and click Install
3. Enter a VNC password when prompted
4. Docker builds the container (takes a few minutes on first install)
5. The gateway restarts — the Browser panel appears in the sidebar

Dashboard Panel

The Browser panel has three tabs:

Status

• Container status (Running/Stopped) with Start/Stop/Restart buttons
• CDP (Chrome DevTools Protocol) connection status
• Live VNC view — embedded iframe showing the browser in real-time
• "Open VNC Viewer" link for a larger view

Sessions

• List of saved browser sessions (cookies + localStorage)
• Restore sessions to resume where you left off

Skills

• Installed automation skills (e.g., FFFF Filing, custom scraping recipes)

MCP Tools

Core Tools

Tool	Description
crow_browser_launch	Connect to browser via CDP, returns VNC URL
crow_browser_status	Container and CDP health check
crow_browser_navigate	Go to URL with stealth scripts
crow_browser_screenshot	Capture page or element as PNG
crow_browser_fill_form	Fill form fields with human-like typing
crow_browser_click	Click with position randomization
crow_browser_evaluate	Run JavaScript in page context
crow_browser_wait_for_user	Pause for human intervention (CAPTCHA, 2FA)
crow_browser_discover_selectors	Find all interactive elements on page
crow_browser_save_session	Save cookies + localStorage to file
crow_browser_load_session	Restore a saved session

Content Extraction Tools

Tool	Description
crow_browser_extract_text	Clean article text via Mozilla Readability
crow_browser_extract_tables	HTML tables to JSON or CSV
crow_browser_extract_links	All links with text and URLs, filterable
crow_browser_scrape	Structured data via CSS selector mapping
crow_browser_paginate	Follow pagination, collect multi-page results
crow_browser_export	Save scraped data as CSV or JSON file
crow_browser_capture_har	Record network requests for API discovery

Workflows

Basic Navigation & Form Filling

1. crow_browser_launch          → connect to browser
2. crow_browser_navigate        → go to the site
3. crow_browser_discover_selectors → find form fields
4. crow_browser_fill_form       → fill in values
5. crow_browser_click           → submit
6. crow_browser_screenshot      → verify result

Content Scraping

1. crow_browser_navigate        → go to the page
2. crow_browser_extract_text    → get clean article text
   — or —
   crow_browser_scrape          → extract structured data via CSS selectors
3. crow_browser_export          → save as CSV or JSON

Multi-Page Scraping

1. crow_browser_navigate        → go to first page
2. crow_browser_paginate        → follow "next" links, extract from each page
3. crow_browser_export          → save combined results

Session Management

1. crow_browser_save_session    → save cookies before long operations
2. (... time passes, session might expire ...)
3. crow_browser_load_session    → restore cookies and continue

Stealth Features

The browser includes anti-detection measures:

• navigator.webdriver masked
• User-Agent rotation — Chrome on Windows, macOS, or Linux profiles
• Plugin spoofing — fake Chrome PDF Plugin, PDF Viewer, Native Client
• Screen dimensions — 1920x1080 with realistic available area
• window.chrome object mocked
• Timezone configurable (defaults to Central)
• Human-like typing — per-character delays with randomization
• Human-like clicking — position randomized within element bounds
• Navigation pauses — random delays between actions

Human Intervention

For CAPTCHA, 2FA, security questions, or any action requiring human judgment:

1. The AI calls crow_browser_wait_for_user with a message
2. You see the message and open the VNC viewer
3. Complete the action manually in the browser
4. Tell the AI to continue (it calls wait_for_user with resume: true)

VNC Access

The VNC viewer is accessible through the Crow gateway:

• Embedded: In the Browser panel's Status tab (iframe)
• Full view: /proxy/browser/vnc.html (same HTTPS as your dashboard)
• Direct: http://localhost:6080/vnc.html (local access only)

No firewall ports need to be opened — VNC is proxied through the gateway.

Docker Container

The container runs:

• Ubuntu 22.04 with Xvfb (virtual framebuffer)
• Playwright Chromium (latest)
• x11vnc + noVNC for browser viewing
• Host networking (network_mode: host) for CDP access

Container limits: 2GB RAM, 1GB shared memory.

Container Management

From the Browser panel:

• Start / Stop / Restart buttons
• Container auto-restarts on reboot (restart: unless-stopped)

From the command line:

cd ~/.crow/bundles/browser
docker compose up -d      # start
docker compose down       # stop
docker compose logs -f    # view logs

FFFF Filing Skill

The Browser Automation extension includes a skill for filing taxes via IRS Free File Fillable Forms. See the Tax Filing Assistant guide for details.

Security

• CDP port (9222) binds to 127.0.0.1 only
• VNC is proxied through the gateway with session authentication
• No ports exposed to the network — everything goes through the gateway's HTTPS
• VNC password is required (set during installation)