kernel-browser-management（内核浏览器管理）

Browser Management

Create and manage sandboxed Chrome browser instances in the cloud.

When to Use This Skill

Use browser-management skill when you need to:

• Create browser sessions - Launch new Chrome browser instances with custom configurations (stealth mode, headless, profiles, proxies)
• List and monitor browsers - View all active browser sessions and their details
• Get live view URLs - Access remote browser sessions for monitoring and control
• Execute automation - Run Playwright/TypeScript code against browser sessions
• Capture screenshots - Take screenshots of browser pages or specific regions
• Manage browser lifecycle - Delete browser sessions when done to free resources
• Work with browser profiles - Load saved authentication data and cookies into sessions

Prerequisites

Load the kernel-cli skill for Kernel CLI installation and authentication.

Create a Browser

# Basic browser creation
kernel browsers create

# With options
kernel browsers create --stealth --headless -o json
kernel browsers create --profile-name my-profile

Output contains session_id, cdp_ws_url, and browser_live_view_url.

MCP Tool: Use kernel:create_browser with parameters like headless, stealth, or profile_name.

List and Get Browsers

Unless otherwise noted, id arguments refer to the browser session ID, not invocation IDs returned by Kernel commands.

kernel browsers list -o json
kernel browsers get <session_id> -o json
kernel browsers view <session_id> -o json    # Get live view URL

MCP Tools: Use kernel:list_browsers, kernel:get_browser.

Delete a Browser

kernel browsers delete <session_id> --yes

MCP Tool: Use kernel:delete_browser with the session_id.

Browser Automation

Execute Playwright Code

Run Playwright/TypeScript code against a browser session:

kernel browsers playwright execute <session_id> 'await page.goto("https://example.com")'

MCP Tool: Use kernel:execute_playwright_code to run automation scripts. If no session_id is provided, a new browser is created and cleaned up automatically.

Take Screenshots

Capture screenshots of browser pages:

kernel browsers computer screenshot <session_id> --to screenshot.png

MCP Tool: Use kernel:take_screenshot with session_id. Optionally specify region with x, y, width, height.

Common Pattern: Create, Use, Delete

# Create browser and capture session_id
SESSION=$(kernel browsers create -o json | jq -r '.session_id')

# Use the browser...
# [perform operations]

# Cleanup
kernel browsers delete $SESSION --yes