Playwright (default)

browser_engine: playwright — the default. Uses the official @playwright/mcp server over stdio. Nothing to install or sign up for; npx fetches it on first use and Playwright manages its own browser binaries.

When to use it

• You want the broadest browser coverage — Chromium, Chrome, Edge, Firefox, WebKit.
• You want full result.md fidelity: embedded screenshots, console errors, and network failures all captured directly.
• You want fully local, isolated processes with zero cloud dependency.

This is the recommended engine for almost everyone.

Configuration

/qa-catalog:init writes this mcpServers block into each project-level browser agent:

mcpServers:
  - playwright:
      type: stdio
      command: npx
      args: ["-y", "@playwright/mcp@latest"]

Each parallel spawn starts its own npx @playwright/mcp@latest process — true OS-process isolation, no shared cookies/localStorage. No browser_new_context call is needed (or exposed); navigate directly.

Relevant settings (/plugin)

Setting	Effect
browser_channel	chromium (default), chrome, msedge, firefox, webkit.
browser_headless	true (default) runs without a visible window; false to watch.
settle_ms	Wait after navigation before snapshotting (slow SPAs).
parallel_agents / parallel_test_runners	Concurrent browser processes. The dev server is the practical ceiling.

Requirements

• Node.js + npx on PATH.
• First run downloads Playwright browser binaries (one-time, automatic).

Notes

• Auth modes (none, shared-credentials, storage-state) are all supported. storage-state reads a Playwright storageState JSON and injects it via browser_evaluate before navigating.
• No account, API key, or network egress to third-party services.