playwright

name	playwright
description	MUST USE for any browser-related tasks. Browser automation via Playwright MCP - verification, browsing, information gathering, web scraping, testing, screenshots, and all browser interactions.
argument-hint	[url or browser task]
user-invocable	true

Playwright Browser Automation

Task: $ARGUMENTS

No task specified → ask user what browser task to perform.

Browser automation via Playwright MCP: testing, scraping, verification, interaction.

Prerequisites

The Playwright MCP server must be configured. Add to your project's .mcp.json:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest"]
    }
  }
}

Or for headed mode (visible browser):

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["@playwright/mcp@latest", "--headless=false"]
    }
  }
}

Available MCP Tools

When the Playwright MCP server is running, these tools become available:

Navigation

• browser_navigate: go to a URL
• browser_go_back / browser_go_forward: browser history navigation
• browser_wait: wait for page load or condition

Interaction

• browser_click: click an element (uses accessibility snapshot selectors)
• browser_type: type text into focused element
• browser_fill: fill form field (clears existing value first)
• browser_select_option: select dropdown option
• browser_hover: hover over element
• browser_drag: drag from one element to another
• browser_press_key: press keyboard key (Enter, Tab, Escape, etc.)

Observation

• browser_snapshot: get accessibility tree snapshot (preferred over screenshot)
• browser_screenshot: take a PNG screenshot
• browser_get_text: extract text content from page
• browser_execute_javascript: run JavaScript in page context

Tab Management

• browser_tab_list: list open tabs
• browser_tab_new: open new tab
• browser_tab_select: switch to tab
• browser_tab_close: close tab

Workflow Patterns

Web Verification

1. browser_navigate to the URL
2. browser_snapshot to get page structure
3. Verify expected elements exist
4. browser_screenshot for visual evidence

Form Testing

1. browser_navigate to form page
2. browser_fill each form field
3. browser_click submit button
4. browser_snapshot to verify result

Web Scraping

1. browser_navigate to target page
2. browser_snapshot for structured content
3. browser_execute_javascript for complex extraction
4. Process results

Multi-Page Flow

1. Navigate to starting page
2. Interact with elements (click links, fill forms)
3. Use browser_snapshot after each navigation to understand new page
4. Continue until flow complete

Fallback

If Playwright MCP server is not configured (no mcp__playwright__* tools available), use /oh-my-claudeagent:dev-browser if its runtime is present. Otherwise ask the user to configure Playwright MCP.

MCP Tool Interaction

Primary interaction is via Playwright MCP tools (mcp__playwright__*), not via Bash. Use the MCP tools directly for all browser operations.

Important Notes

• Prefer browser_snapshot over browser_screenshot (structured accessibility data)
• Element selectors from browser_snapshot output
• One session persists across calls
• Element not found → new snapshot to see current state