playread

Google search automation via Playwright MCP. Can be used as a CLI tool or as an MCP server.

Installation

npm install -g playread

Or use directly with npx:

npx -y playread@latest google-search

Usage

As an MCP Server

Start playread as an MCP server to expose all flows as tools:

playread mcp

Or configure in your MCP client settings (e.g., Claude Desktop):

{
  "mcpServers": {
    "playread": {
      "command": "npx",
      "args": ["-y", "playread@latest", "mcp"]
    }
  }
}

Available MCP tools:

• fetch - Fetch and extract main content from a web page URL
• google-search - Perform a Google search and extract results

As a CLI Tool

Google Search

playread google-search

Output format:

{
  "query": "Playwright automation testing",
  "totalResults": 18,
  "results": [
    {
      "title": "Playwright: Fast and reliable end-to-end testing for modern ...",
      "url": "https://playwright.dev/",
      "description": "Playwright enables reliable end-to-end testing for modern web apps..."
    }
  ]
}

Fetch URL Content

Extract important text content from any URL:

playread fetch https://example.com

Output format:

{
  "url": "https://example.com/",
  "title": "Example Domain",
  "contentLength": 129,
  "content": "# Example Domain\nThis domain is for use in documentation examples..."
}

Perfect for:

• Extracting documentation from websites
• Fetching article content
• Parsing blog posts
• Getting README content from GitHub
• Any text-heavy webpage

Creating Custom Flows

Flows are JavaScript modules that export a single async function. The function receives a connected PlaywrightMCPClient instance.

Create a new flow in the flows/ directory:

module.exports = async function(client) {
  await client.navigate('https://example.com');

  const snapshot = await client.snapshot();
  console.log('Page loaded');

  await client.close();
};

Available Client Methods

Connection

• connect() - Connect to Playwright MCP server
• disconnect() - Disconnect and cleanup

Tools Discovery

• listTools() - List all available tools
• callTool(name, args) - Call any tool directly

Browser Control

• navigate(url) - Navigate to URL
• navigateBack() - Go back
• close() - Close browser
• resize(width, height) - Resize window
• install() - Install browser

Page Interaction

• snapshot() - Get accessibility snapshot
• click(element, ref) - Click element
• type(element, ref, text, submit, slowly) - Type text
• pressKey(key) - Press keyboard key
• hover(element, ref) - Hover over element
• drag(startElement, startRef, endElement, endRef) - Drag and drop

Forms

• fillForm(fields) - Fill multiple form fields
• selectOption(element, ref, values) - Select dropdown option
• fileUpload(paths) - Upload files

Page State

• evaluate(func) - Execute JavaScript
• screenshot(filename, type, fullPage, element, ref) - Take screenshot
• consoleMessages(onlyErrors) - Get console logs
• networkRequests() - Get network activity

Dialogs

• handleDialog(accept, promptText) - Handle alerts/confirms

Tabs

• tabs(action, index) - Manage browser tabs

Wait

• waitFor(text, textGone, time) - Wait for conditions

Flow Details

google-search

• Navigates to Google
• Searches for "Playwright automation testing"
• Waits 30 seconds for results to load
• Extracts all result titles, URLs, and descriptions
• Returns comprehensive JSON output with full result metadata

fetch

• Accepts a URL as an argument
• Navigates to the URL
• Intelligently extracts main content using heuristics:
  • Prioritizes <main>, <article>, and semantic content areas
  • Extracts structured text including headings, paragraphs, lists, and code blocks
  • Removes duplicate content and navigation elements
  • Formats output as clean markdown
• Returns JSON with URL, title, content length, and extracted content
• Limits output to 50,000 characters for conciseness
• Works universally across documentation sites, blogs, articles, and more

Requirements

• Node.js 14+
• Playwright MCP server (installed automatically via npx)

Notes

• Always close the browser before flows complete
• Supports file:// URLs for local testing
• Use browser_evaluate for debugging window globals
• No hardcoded values - all flows use ground truth from snapshots
• Screenshots are automatically saved to /tmp/playwright-mcp-output/<timestamp>/