Browser
The browser tool gives agents full control over a real web browser using Playwright. Agents can navigate to URLs, click elements, fill forms, take screenshots, and extract content — making it possible to interact with any website including JavaScript-heavy SPAs.
Tools
| Tool | Description |
|---|---|
browser_navigate | Navigate to a URL and return the page content |
browser_click | Click on an element (by selector or text) |
browser_fill | Fill an input field |
browser_screenshot | Take a screenshot and return it as base64 |
browser_extract | Extract specific elements from the current page |
Use cases
- Research — browse URLs, read articles, follow links
- Form automation — log in, fill forms, submit data
- Web monitoring — screenshot pages for visual diffs
- Data extraction — extract structured data from dynamic pages (consider the web scraper for heavy crawling)
- Testing — automated UI testing scenarios
Example agent interactions
Go to https://example.com/login, log in with username "admin" and password from vault/creds.txt, then navigate to the dashboard and screenshot it.
Search for "Synapse AI" on GitHub and return the top 5 repository results with their star counts.
Go to our internal Jira board at https://jira.internal/board and list all tickets in the "In Progress" column.
Playwright configuration
The browser runs in headless mode by default. It uses Chromium. No additional configuration is required — Playwright is installed as part of the backend dependencies.
If Playwright browsers are not installed:
npx @playwright/mcp install-browser chrome-for-testing
Comparison with Web Scraper
| Browser | Web Scraper | |
|---|---|---|
| JavaScript rendering | Yes | Yes (crawl4ai) |
| Anti-bot evasion | Basic | Advanced (stealth mode) |
| Interactive forms | Yes | No |
| Screenshots | Yes | No |
| Multi-page crawling | No (manual) | Yes (automatic) |
| Best for | Interactive sites, forms, SPAs | High-volume scraping, paywalled sites |