🎭 MCP Playwright Server

MCP (Model Context Protocol) server that enables Claude Code to control a browser via Playwright. Perfect for web scraping, automation, and research in research-modular.

Features

• ✅ Navigate: Load any website and wait for dynamic content

• ✅ Screenshot: Capture full pages or viewport screenshots

• ✅ Scrape: Extract text, links, images, forms with structured output

• ✅ Form Fill: Interact with forms, fill fields, submit

• ✅ PDF Export: Generate PDFs from any page

• ✅ Session Management: Multiple browser sessions simultaneously

• ✅ Output Formats: JSON, Markdown, HTML

Related MCP server: Playwright Plus Python MCP

Installation

Option 1: npm (once published)

npm install -g mcp-playwright-server

Option 2: Direct from GitHub (now)

npx -y github:Jens0110/mcp-playwright-server

Option 3: Local development

git clone https://github.com/Jens0110/mcp-playwright-server
cd mcp-playwright-server
npm install
npm run build
npm start

Setup with Claude Code

1. Add MCP server to Claude config

claude mcp add playwright -- npx -y github:Jens0110/mcp-playwright-server

Or manually in ~/.claude/config.json:

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": ["-y", "github:Jens0110/mcp-playwright-server"]
    }
  }
}

2. Restart Claude Code

The browser automation tools are now available in Claude Code.

Usage Examples

Navigate to a website

use_mcp_tool: playwright_navigate
{
  "url": "https://www.fillout.com",
  "timeout": 30000
}

Take a screenshot

use_mcp_tool: playwright_screenshot
{
  "session_id": "default",
  "full_page": true,
  "filename": "fillout-homepage.png"
}

Scrape content

use_mcp_tool: playwright_scrape
{
  "session_id": "default",
  "extract_text": true,
  "extract_links": true,
  "extract_forms": true,
  "format": "json"
}

Fill and submit a form

use_mcp_tool: playwright_form_fill
{
  "session_id": "default",
  "selectors": {
    "input[name='email']": "user@example.com",
    "input[name='password']": "secret123",
    "input[type='checkbox']": "on"
  },
  "submit_selector": "button[type='submit']",
  "wait_for_navigation": true
}

Generate PDF

use_mcp_tool: playwright_pdf
{
  "session_id": "default",
  "filename": "page-export.pdf",
  "format": "A4"
}

Integration with research-modular

Perfect for competitive analysis:

/research-modular

"Analyze Fillout's pricing page"

→ navigate to fillout.com/pricing
→ screenshot
→ scrape pricing tables
→ extract feature comparison
→ → Add to Feature-Matrix
→ → Generate Positioning Map

Output Locations

• Screenshots: ~/.cache/research-modular/screenshots/

• PDFs: ~/.cache/research-modular/pdfs/

• Scraped data: JSON output (in-memory)

Session Management

Each session maintains its own browser context:

// Session 1: Analyze Competitor A
playwright_navigate({ url: "competitor-a.com", session_id: "competitor-a" })
playwright_scrape({ session_id: "competitor-a" })

// Session 2: Analyze Competitor B (parallel!)
playwright_navigate({ url: "competitor-b.com", session_id: "competitor-b" })
playwright_scrape({ session_id: "competitor-b" })

Performance

• First page load: ~3-5 seconds (browser startup)

• Subsequent pages: ~1-2 seconds

• Screenshot generation: ~500ms

• PDF generation: ~1-2 seconds

• Memory: ~200MB per active session

Limitations

• No support for authenticated browsing yet (Session 2)

• JavaScript execution limited to DOM manipulation (no custom scripts)

• Form interactions limited to standard inputs (no complex JS-heavy forms)

Troubleshooting

"Browser launch failed"

# Install browser dependencies
npm install -D @playwright/test
npx playwright install chromium

"Connection timeout"

• Increase timeout parameter (in milliseconds)

• Check internet connection

• Website might be blocking automated access

"Form fill not working"

• Verify CSS selector is correct

• Try waiting longer with navigation

• Check if form requires JavaScript initialization

Future Enhancements

• Session persistence (cookies/auth)

• Custom JavaScript execution

• Network interception & HAR export

• Performance metrics collection

• Video recording

• Mobile device emulation

Architecture

Claude Code
    ↓
[MCP Protocol]
    ↓
Playwright Server
    ↓
[Browser Pool]
    ↓
Chromium Browser
    ↓
[websites]

Contributing

Contributions welcome! Please:

1. Fork the repo

2. Create a feature branch

3. Add tests

4. Submit PR

License

MIT

Support

• Issues: https://github.com/Jens0110/mcp-playwright-server/issues

• Docs: https://playwright.dev

• MCP Spec: https://modelcontextprotocol.io