Browser Automation & Test Automation

Four approaches for automating browser operations and testing with Claude Code.

Which One Should You Start With?

Option 1: Claude in Chrome

Chrome extension. Claude assists while seeing your screen. Install and go — no setup needed.

• Review test procedures together visually
• Ask “Is this page displaying correctly?”
• Accessible for non-engineers

Option 2: Cowork

Visit claude.ai/cowork  for screen sharing with Claude. No installation needed.

• Give instructions like “click here on this screen”
• Show your workflow: “I do this every time. Can it be automated?”

Option 3: Playwright MCP

Claude Code directly controls the browser. Best for automating repetitive tests.

Setup

Add to .mcp.json:

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

Works immediately if Node.js is installed.

Usage

"Open this URL, log in, and screenshot the order list"
"Fill this form with 10 test entries"
"Navigate product → cart → checkout, screenshot each step"

Playwright CLI (Token-Efficient)

@playwright/cli uses shell commands instead of MCP. ~4x more token-efficient (114k → 27k tokens/task).

npm install -g @playwright/cli

Option 4: Playwright Test Agents

Microsoft’s AI test agents covering the full cycle: planning → code generation → healing.

Setup

npx playwright init-agents

Always review auto-generated test plans and test code to ensure they match your intended test scenarios before running them.

Before/After

References

• Playwright MCP (Microsoft Official) 
• Playwright CLI (Token-Efficient) 
• Playwright Test Agents 
• Playwright Official