browser-use

mjunaidca 的头像作者 mjunaidca

使用Playwright MCP进行浏览器自动化。导航网站、填写表单、点击元素、截图和提取数据。适用于需要网络浏览、表单提交、网络抓取、UI测试或任何浏览器交互的任务。

📖 中文导读:使用Playwright MCP进行浏览器自动化。导航网站、填写表单、点击元素、截图和提取数据。适用于需要网络浏览、表单提交、网络抓取、UI测试或任何浏览器交互的任务。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: browser-use description: Browser automation using Playwright MCP. Navigate websites, fill forms, click elements, take screenshots, and extract data. Use when tasks require web browsing, form submission, web scraping, UI testing, or any browser interaction.

Browser Automation

Automate browser interactions via Playwright MCP server.

Server Lifecycle

Start Server

# Using helper script (recommended)
bash scripts/start-server.sh

# Or manually
npx @playwright/mcp@latest --port 8808 --shared-browser-context &

Stop Server

# Using helper script (closes browser first)
bash scripts/stop-server.sh

# Or manually
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_close -p '{}'
pkill -f "@playwright/mcp"

When to Stop

  • End of task: Stop when browser work is complete
  • Long sessions: Keep running if doing multiple browser tasks
  • Errors: Stop and restart if browser becomes unresponsive

Important: The --shared-browser-context flag is required to maintain browser state across multiple mcp-client.py calls. Without it, each call gets a fresh browser context.

Quick Reference

# Go to URL
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_navigate \
  -p '{"url": "https://example.com"}'

# Go back
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_navigate_back -p '{}'

Get Page State

# Accessibility snapshot (returns element refs for clicking/typing)
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_snapshot -p '{}'

# Screenshot
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_take_screenshot \
  -p '{"type": "png", "fullPage": true}'

Interact with Elements

Use ref from snapshot output to target elements:

# Click element
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_click \
  -p '{"element": "Submit button", "ref": "e42"}'

# Type text
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_type \
  -p '{"element": "Search input", "ref": "e15", "text": "hello world", "submit": true}'

# Fill form (multiple fields)
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_fill_form \
  -p '{"fields": [{"ref": "e10", "value": "john@example.com"}, {"ref": "e12", "value": "password123"}]}'

# Select dropdown
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_select_option \
  -p '{"element": "Country dropdown", "ref": "e20", "values": ["US"]}'

Wait for Conditions

# Wait for text to appear
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_wait_for \
  -p '{"text": "Success"}'

# Wait for time (ms)
python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_wait_for \
  -p '{"time": 2000}'

Execute JavaScript

python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_evaluate \
  -p '{"function": "return document.title"}'

Multi-Step Playwright Code

For complex workflows, use browser_run_code to run multiple actions in one call:

python3 scripts/mcp-client.py call -u http://localhost:8808 -t browser_run_code \
  -p '{"code": "async (page) => { await page.goto(\"https://example.com\"); await page.click(\"text=Learn more\"); return await page.title(); }"}'

Tip: Use browser_run_code for complex multi-step operations that should be atomic (all-or-nothing).

Workflow: Form Submission

  1. Navigate to page
  2. Get snapshot to find element refs
  3. Fill form fields using refs
  4. Click submit
  5. Wait for confirmation
  6. Screenshot result

Workflow: Data Extraction

  1. Navigate to page
  2. Get snapshot (contains text content)
  3. Use browser_evaluate for complex extraction
  4. Process results

Tool Reference

See references/playwright-tools.md for complete tool documentation.

Troubleshooting

Issue Solution
Element not found Run browser_snapshot first to get current refs
Click fails Try browser_hover first, then click
Form not submitting Use "submit": true with browser_type
Page not loading Increase wait time or use browser_wait_for