browser

name	browser
description	Browse the web using a real headed Chromium browser on a virtual desktop. Supports navigation, clicking, typing, screenshots, and full page interaction. The user can watch the browser in real-time via noVNC.

Browser Skill

You have access to a virtual desktop with a headed Chromium browser for web browsing, automation, and interaction. This is a REAL browser (not headless), which means it works with sites that detect and block headless browsers.

Quick Start

Step 1: Launch the virtual display

Call the tower_display_launch tool (no arguments). This starts the virtual desktop that the browser renders on. It returns a noVNC URL the user can open to watch.

Step 2: Launch Chromium

Call the tower_display_browser tool. This starts Chromium on the display with a per-session profile (logins and cookies persist across sessions). You can optionally pass a url parameter to open a specific page.

The tool returns:

• cdpPort — Chrome DevTools Protocol port for direct browser control
• chromiumProfileDir — per-session profile directory
• noVncUrl — URL for the user to watch

Step 3: Interact with the page

Primary method — Playwright MCP tools (fastest, most reliable):

• browser_navigate(url) — go to a URL
• browser_click(element) — click an element
• browser_type(element, text) — type into an input
• browser_snapshot — get the accessibility tree (structured page content)
• browser_take_screenshot — capture the browser viewport

Alternative — CDP direct control (for advanced use cases):

List open tabs:

curl -s http://localhost:<cdpPort>/json/list

Navigate via CDP:

curl -s http://localhost:<cdpPort>/json/list | python3 -c "
import sys, json, urllib.request
tabs = json.load(sys.stdin)
tab_id = tabs[0]['id']
urllib.request.urlopen(f'http://localhost:<cdpPort>/json/navigate/{tab_id}?url=https://example.com')
print('Navigated')
"

Alternative — Computer-use tools (for visual interaction):

• screenshot — full display screenshot
• mouse_click(x, y) — click at pixel coordinates
• type_text(text) — type text
• press_key(keys) — press key combinations

Available tools

• tower_display_launch — start the virtual desktop (MUST call first)
• tower_display_browser — start Chromium on the display
• tower_display_terminal — start a terminal emulator on the display
• tower_display_status — check if a display is running
• tower_display_destroy — shut down the display and all apps on it
• Playwright MCP tools — browser_navigate, browser_click, browser_snapshot, etc.
• Computer-use tools — screenshot, mouse_click, type_text, press_key, etc.

Tips

• The user can watch what you're doing in real-time by opening the noVNC URL returned by tower_display_launch.
• Prefer Playwright MCP tools for structured web interaction — they're faster and more reliable than screenshot-based approaches.
• Use computer-use tools when you need to interact with non-browser elements (e.g., browser dialogs, desktop notifications) or when Playwright can't reach what you need.
• Use xdotool via bash for keyboard/mouse input outside of both Playwright and computer-use tools.
• Always check if the browser is running with ps aux | grep chromium before trying to interact with it.
• If the browser crashes, call tower_display_browser again to relaunch it.