| name | Browser Automation |
| description | Automate browser interactions via Playwright MCP. Screenshots, scraping, cookie management, form filling, and visual testing.
<example>
Context: User wants to capture a webpage
user: "Take a screenshot of example.com"
</example>
<example>
Context: User wants to interact with a page
user: "Go to github.com and click the sign in button"
</example>
<example>
Context: User needs to manage browser state
user: "Save the current browser session so I can resume later"
</example>
<example>
Context: User wants to extract data
user: "Query all the links on this page"
</example>
|
| version | 1.0.0 |
Browser Automation
Automate browser interactions using Playwright via the browse MCP server.
Full tool reference: references/tools.md
Session Startup (IMPORTANT)
On first browse tool use each session, restore saved cookies:
mcp__browse__session_restore(path: "/Users/chi/.claude/browse-session.json")
This restores 400+ authenticated cookies from Safari. Skip only if:
- File doesn't exist
- User explicitly wants fresh session
- Already restored this session
Before ending session, save if cookies changed:
mcp__browse__session_save(path: "/Users/chi/.claude/browse-session.json")
Quick Start
Basic Screenshot
1. mcp__browse__goto(url: "https://example.com")
2. mcp__browse__screenshot()
Interactive Session
1. mcp__browse__launch(headed: true)
2. mcp__browse__goto(url: "https://github.com")
3. mcp__browse__click(selector: "a[href='/login']")
4. mcp__browse__type(selector: "#login_field", text: "username")
Query Elements
1. mcp__browse__goto(url: "https://example.com")
2. mcp__browse__query(selector: "a[href]") # Get all links
Browser Modes
| Mode | Launch | Use Case |
|---|
| Headless | launch() or omit | Default, fastest |
| Headed | launch(headed: true) | Debugging |
| Fullscreen | launch(fullscreen: true) | macOS native fullscreen |
| Preview | launch(preview: true, previewDelay: 2000) | Highlight before action |
Session Persistence
Save: mcp__browse__session_save(path: "my-session.json") — saves URL, cookies, localStorage, sessionStorage.
Restore: mcp__browse__session_restore(path: "my-session.json")
Import Safari cookies (macOS, requires Full Disk Access):
mcp__browse__import(domain: "github.com") # Specific domain
mcp__browse__import() # All cookies
Common Patterns
Login and Save Session
1. launch(headed: true)
2. goto(url: "https://site.com/login")
3. type(selector: "#email", text: "user@example.com")
4. type(selector: "#password", text: "secret")
5. click(selector: "button[type='submit']")
6. wait(ms: 3000)
7. session_save(path: "site-session.json")
8. close()
Scrape Data
1. goto(url: "https://example.com/products")
2. query(selector: ".product-card")
Best Practices
- Launch once per session — reuse the browser instance
- Wait after navigation — pages need time to load
- Use specific selectors — avoid ambiguous matches
- Save sessions for auth — don't re-login every time
- Check for errors — use
errors() and console() to debug
- Close when done — free resources with
close()
