playwright

어떤 스킬인가요?

Browser automation via Playwright MCP. Use for web scraping, taking screenshots, interacting with web pages, testing web UIs, and automating browser tasks. Headless browser support.

설치 방법

프로젝트 폴더에서 아래 명령어를 실행하고, 설치할 에이전트를 선택하세요.

설치 명령을 직접 실행해야 적용됩니다. 지원 에이전트와 필요한 권한·라이선스는 제작자의 안내를 확인하세요.

스킬 원문 파일

<!-- BEGIN GENERATED related (scripts/relate_skills.py) -->

## Related

Skills related to this one, with the nature of each relation (direction from prose citations; cluster domains characterized from exa and our own use). Symmetric — each related skill lists this one too. Do not edit inside the markers; regenerate with `python3 scripts/relate_skills.py`.

- `amazon-shopping-mcp` — sibling in the skill routing & dispatch cluster — Amazon product search, cart management, price tracking, and shopping list via Playwright …
- `amp-team-usage` — invoked by — Amp Team Usage
- `browser-navigation` — sibling in the skill routing & dispatch cluster — |
- `develop-web-game` — invoked by — Use when Codex is building or iterating on a web game (HTML JS) and needs a reliable deve…
- `find-skills` — sibling in the skill routing & dispatch cluster — Helps users discover and install agent skills when they ask questions like "how do I do X…
- `partiful` — invoked by — Access Partiful events, invites, and RSVPs via reverse engineered API.
- `playwright-unworld` — sibling in the skill routing & dispatch cluster — Playwright Unworld Skill: Deterministic Web Automation
- `qa-regression` — sibling in the skill routing & dispatch cluster — Automate QA regression testing with reusable test skills.
- `ruler-maximal` — sibling in the skill routing & dispatch cluster — Ruler Maximal Skill
- `screenshot` — sibling in the skill routing & dispatch cluster — Use when the user explicitly asks for a desktop or system screenshot (full screen, specif…
- `sparc-methodology` — sibling in the skill routing & dispatch cluster — SPARC (Specification, Pseudocode, Architecture, Refinement, Completion) comprehensive dev…
- `uv-oneliners` — sibling in the skill routing & dispatch cluster — UV UVX awesome one liners for ephemeral Python environments with multi package
- `webapp-testing` — invoked by — Toolkit for interacting with and testing local web applications using

<!-- END GENERATED related -->

---
name: playwright
description: Browser automation via Playwright MCP. Use for web scraping, taking screenshots, interacting with web pages, testing web UIs, and automating browser tasks. Headless browser support.
compatibility: Requires npx and @anthropic-ai/mcp-server-playwright package.
---

# Playwright Browser Automation

Control browsers via Playwright MCP server.

## When to Use

- Web scraping and data extraction
- Taking screenshots of web pages
- Interacting with web UIs (clicking, typing, navigating)
- Testing web applications
- Automating browser-based workflows
- Filling forms and submitting data

## Setup

MCP server configured in `~/.mcp.json`:
```json
{
  "playwright": {
    "command": "npx",
    "args": ["-y", "@anthropic-ai/mcp-server-playwright"]
  }
}
```

## Common Tools

### Navigation
- `navigate_page` - Go to a URL
- `new_page` - Open new browser tab
- `list_pages` - Show open pages

### Interaction
- `click` - Click elements
- `fill` - Type into input fields
- `select` - Choose from dropdowns
- `press` - Press keyboard keys

### Capture
- `take_screenshot` - Screenshot current page
- `get_page_content` - Get page HTML
- `get_text` - Extract visible text

### Evaluation
- `evaluate` - Run JavaScript in page context

## Example Workflows

### Screenshot a Page
1. `navigate_page(url="https://example.com")`
2. `take_screenshot()`

### Fill a Form
1. `navigate_page(url="https://example.com/form")`
2. `fill(selector="#email", value="user@example.com")`
3. `fill(selector="#password", value="secret")`
4. `click(selector="button[type=submit]")`

### Extract Data
1. `navigate_page(url="https://example.com/data")`
2. `get_text(selector=".results")`

## Tips

- Use CSS selectors or XPath for element targeting
- Wait for page loads before interacting
- Browser runs headless by default
- Screenshots are useful for debugging