browser-automation

Browser automation via Puppeteer MCP for JS-rendered content

⭐2stars🔀0forks📁View on GitHub🕐Updated Jan 3, 2026⬇️3downloads

When & Why to Use This Skill

This Claude skill enables advanced web automation and headless browser control using Playwright and Puppeteer MCP servers. It allows Claude to interact with modern, JavaScript-heavy websites, capture screenshots, fill forms, and extract data from dynamic pages that traditional static scrapers cannot handle.

Use Cases

  • Extracting data from dynamic Single Page Applications (SPAs) built with frameworks like React, Vue, or Angular where content is rendered via JavaScript.
  • Automating multi-step web workflows, such as logging into a secure portal, navigating through dashboards, and retrieving specific reports.
  • Capturing visual evidence through full-page or element-specific screenshots for UI/UX auditing, bug reporting, or monitoring visual changes.
  • Performing automated web testing by simulating user interactions like clicking buttons, filling out complex forms, and hovering over elements.
  • Accessing and scraping content from authentication-protected websites by maintaining browser sessions and managing cookies.
namebrowser-automation
descriptionBrowser automation via Puppeteer MCP for JS-rendered content

Browser Automation

Use Playwright or Puppeteer MCP for browser automation when WebFetch isn't sufficient.

Overview

Browser automation provides headless browser control for:

  • JS-rendered content extraction
  • Screenshot capture
  • Form filling and submission
  • Interactive testing
  • Auth-protected page access

MCP Servers

Multi-browser support: Firefox, Chrome, WebKit. Same API across all browsers.

Package: @anthropic/mcp-server-playwright Transport: stdio Install: claude mcp add playwright -- npx -y @anthropic/mcp-server-playwright

# Also install browser binaries
npx playwright install firefox chromium webkit

Puppeteer (Chrome Only)

Package: puppeteer-mcp-server Transport: stdio Install: claude mcp add puppeteer -- npx -y puppeteer-mcp-server

Comparison

Feature Playwright Puppeteer
Firefox ✓ ✗
Chrome/Chromium ✓ ✓
WebKit/Safari ✓ ✗
API Same across browsers Chrome-specific
Maintenance Microsoft (active) Google

Available Tools

puppeteer_navigate Navigate to a URL and wait for page load.

url: "https://example.com"
waitUntil: "networkidle0" | "domcontentloaded" | "load"

Screenshots

puppeteer_screenshot Capture the current page or specific element.

name: "screenshot-name"
selector: "#element-id"  (optional, full page if omitted)
fullPage: true  (optional)

Interaction

puppeteer_click Click an element by CSS selector.

selector: "button.submit"

puppeteer_fill Fill an input field with text.

selector: "input[name=email]"
value: "user@example.com"

puppeteer_select Select an option from a dropdown.

selector: "select#country"
value: "US"

puppeteer_hover Hover over an element (for tooltips, dropdowns).

selector: ".menu-trigger"

JavaScript Execution

puppeteer_evaluate Run JavaScript in the page context and return results.

script: "document.title"

Usage Patterns

Extract JS-Rendered Content

1. puppeteer_navigate(url="https://spa-app.com/data")
2. puppeteer_evaluate(script="JSON.stringify(window.__DATA__)")

Fill and Submit Form

1. puppeteer_navigate(url="https://example.com/login")
2. puppeteer_screenshot(name="before-login")
3. puppeteer_fill(selector="input[name=email]", value="user@example.com")
4. puppeteer_fill(selector="input[name=password]", value="secret")
5. puppeteer_click(selector="button[type=submit]")
6. puppeteer_screenshot(name="after-login")

Capture Visual Evidence

1. puppeteer_navigate(url="https://example.com")
2. puppeteer_screenshot(name="homepage", fullPage=true)
3. puppeteer_click(selector=".open-modal")
4. puppeteer_screenshot(name="modal-open")

Decision Tree

Need web content?
    │
    ├─ Static HTML? ──────────────────→ WebFetch
    │
    ├─ JS-rendered (React, Vue, etc.)? → Puppeteer
    │
    ├─ Need screenshot? ──────────────→ Puppeteer
    │
    ├─ Need to fill forms? ───────────→ Puppeteer
    │
    ├─ Auth-protected? ───────────────→ Puppeteer (can maintain session)
    │
    └─ Default ───────────────────────→ WebFetch (simpler, faster)

Error Handling

Element Not Found

If a selector doesn't match:

  1. Take screenshot to see current page state
  2. Use puppeteer_evaluate to check DOM
  3. Try alternative selectors
  4. Add wait time for dynamic content

Timeout

If page takes too long:

  1. Try waitUntil: "domcontentloaded" instead of networkidle0
  2. Check if site blocks automation
  3. Look for loading spinners/skeleton screens

Session Management

  • Browser sessions persist within the conversation
  • Use for multi-step flows (login → navigate → action)
  • Session cleared on conversation end

Comparison with WebFetch

Feature WebFetch Playwright Puppeteer
Speed Fast Slower Slower
JS execution No Yes Yes
Screenshots No Yes Yes
Form filling No Yes Yes
Session/cookies No Yes Yes
Resource usage Low Higher Higher
Firefox N/A Yes No
WebKit/Safari N/A Yes No

Best Practices

  1. Screenshot liberally - Visual evidence helps debugging
  2. Use specific selectors - IDs are most reliable
  3. Wait for content - Dynamic sites need time to load
  4. Handle failures gracefully - Not all elements exist
  5. Close when done - Don't leave orphan sessions

Troubleshooting

"Element not found"

  • Page may not be fully loaded
  • Selector may be wrong
  • Element may be in iframe
  • Site may be slow
  • Site may block headless browsers
  • Network issues

"Click had no effect"

  • Element may be covered by overlay
  • May need to scroll into view
  • May need to wait for animation

Should NOT Attempt

  • Scraping content that's available via API (use the API instead)
  • Bypassing authentication without permission
  • High-frequency automated requests (respect rate limits)
  • Accessing sites that explicitly block automation
  • Using for tasks WebFetch can handle (unnecessary overhead)

Escalation

  • Complex multi-page flows → Consider breaking into smaller steps with verification
  • Site blocking automation → May need proxy or different approach
  • Performance-critical scraping → Consider dedicated scraping service

When Blocked

If browser automation fails:

  1. Take screenshot to diagnose current state
  2. Try alternative selectors or waiting strategies
  3. Check if site blocks headless browsers
  4. Fall back to WebFetch for static content
  5. Report the specific error and page state
  • mcp-cli: Test Puppeteer/Playwright MCP servers before adding to config
  • systematic-debugging: Debug complex browser automation failures