HN 표시: Brw – Claude Code 상담원 팀을 위한 브라우저 자동화

hackernews | | 📰 뉴스
#ai 에이전트 #chrome #claude #claude code #cli #기타 ai #브라우저 자동화
원문 출처: hackernews · Genesis Park에서 요약 및 분석

요약

Brw는 Claude 코드 에이전트 팀을 위한 브라우저 자동화 도구입니다. 이 도구는 Claude가 웹 페이지에서 정보를 효율적으로 추출하고 활용할 수 있도록 설계되었습니다. 특히, 코드 생성 및 디버깅과 같은 작업에서 브라우저 상호작용을 자동화하여 생산성을 높이는 데 초점을 맞추고 있습니다. 현재 베타 버전이며, 관심 있는 사용자는 GitHub에서 자세한 정보를 확인할 수 있습니다.

본문

Browser automation plugin for Claude Code. Control a real Chrome browser via CLI commands — click, type, navigate, screenshot, read pages, and more. output_recording.mp4 - Open and transparent: Claude for Chrome is a black box requiring a subscription. brw is open source with full visibility into what's happening. - Agent-friendly architecture: Playwright MCP and Chrome DevTools MCP servers weren't designed for parallel agent workflows — they struggle with multiple agents sharing one browser. brw uses a proxy with per-tab mutexes, stateless CLI commands, and structured JSON output built for concurrent agent access. - Lightweight: No heavy MCP server overhead. A single proxy manages Chrome, and each CLI call is a simple HTTP request. Gives Claude Code agents the ability to interact with web browsers through a CLI tool (brw ) backed by Chrome DevTools Protocol. A proxy server manages the Chrome instance and handles concurrent access from multiple agents. Capabilities: - Screenshots (viewport, full-page, element-level), clicks, typing, keyboard shortcuts, scrolling, dragging - Page accessibility tree reading with element refs and text search filtering - Form filling (by ref or CSS selector), text extraction, JavaScript execution - Conditional waiting ( wait-for selector/text/URL/JS condition) - Tab management (create, switch, close, list) - Iframe targeting for read-page, JS execution, and form input - Browser dialog handling (alert, confirm, prompt) with auto-dismiss - Console and network monitoring, response body inspection - Network request interception and mocking - Cookie and localStorage/sessionStorage management - GIF recording of browser actions with click/drag overlays - Device/viewport emulation, geolocation, timezone, dark mode - PDF export, performance metrics, download tracking - Quick mode for batching multiple actions in one call - URL allowlisting for restricting navigation scope - JSON output by default with page fingerprinting for navigation detection - Node.js 18+ - A Chromium-based browser (Chrome, Chromium, Edge, or Brave) # Add the marketplace (if not already added) /plugin marketplace add sshh12/claude-plugins # Install the plugin /plugin install brw@shrivu-plugins claude --plugin-dir ./plugins/brw Once installed, Claude will automatically use brw when you ask it to interact with websites. You can also invoke the skill directly: /brw:brw - "Go to example.com and take a screenshot" - "Fill out the login form on localhost:3000 with test credentials" - "Navigate to our staging app and check if the signup flow works" - "Record a GIF of the checkout process on our dev server" - "Test the mobile layout of our landing page" Configuration is resolved in priority order: env vars > .claude/brw.json (repo-local) > ~/.config/brw/config.json (user) > defaults. | Variable | Default | Description | |---|---|---| BRW_PORT | 9225 | Proxy server port | BRW_CDP_PORT | 9222 | Chrome debugging port | BRW_DATA_DIR | ~/.config/brw/chrome-data | Chrome profile directory | BRW_CHROME_PATH | Auto-detect | Path to browser binary | BRW_HEADLESS | false | Run headless (no visible window) | BRW_SCREENSHOT_DIR | /tmp/brw-screenshots | Screenshot output directory | BRW_ALLOWED_URLS | * | Comma-separated URL glob patterns | Restrict agents to your dev server and configure the browser per-project: { "allowedUrls": ["http://localhost:*", "https://staging.myapp.com/*"], "chromeDataDir": "./.chrome-data", "headless": true } brw config Shows every resolved config value and where it came from (env, repo config, user config, or default). Claude Agent ──HTTP──→ Proxy Server ──CDP/WS──→ Chrome (localhost:9225) (localhost:9222) - Proxy server: Auto-launches on first CLI call. Manages Chrome lifecycle, CDP connections, tab state, and per-tab mutexes for safe concurrent access. - CLI ( brw ): Stateless — each call sends an HTTP request to the proxy and prints the result. Mutation commands auto-return a screenshot. - Multi-agent: Multiple agents share one Chrome/proxy instance, isolated by tabs.

Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.

공유

관련 저널 읽기

전체 보기 →