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

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

1. Start Chrome with debugging:
   chrome.exe --remote-debugging-port=9222

2. Log in manually in Chrome

3. In your AI agent:
   → browser_connect()                          # Extract auth from Chrome
   → instance_create({ instanceId: "task-1", url: "https://app.example.com" })
   → instance_create({ instanceId: "task-2", url: "https://app.example.com" })
   → (Both instances are logged in automatically!)

4. Run tasks in parallel:
   → page_browser_click({ instanceId: "task-1", ... })
   → page_browser_click({ instanceId: "task-2", ... })

5. Export auth from a running instance (e.g., after login flow):
   → instance_export_auth({ instanceId: "task-1" })
   → instance_create({ instanceId: "task-3" })  # Also logged in

→ instance_create({ instanceId: "login-bot", url: "https://example.com/login" })
→ page_browser_click({ instanceId: "login-bot", ... })   # complete login
→ instance_export_auth({ instanceId: "login-bot" })      # export & set as global auth
→ instance_create({ instanceId: "worker-1" })            # inherits login state
→ instance_create({ instanceId: "worker-2" })            # inherits login state

npx playwright-mcp-parallel@latest [options]

playwright-mcp-parallel
├── Management Layer
│   ├── browser_connect      → Extract auth from existing Chrome
│   ├── instance_create      → Launch isolated browser + clone auth
│   ├── instance_list        → List active instances
│   ├── instance_close       → Dispose instance
│   ├── instance_close_all   → Dispose all
│   └── instance_export_auth → Export cookies/localStorage from instance
│
└── Per-Instance Tool Dispatch
    └── page_* tools → routed to the correct BrowserBackend by instanceId