agent-browser

name	agent-browser
description	Drive a real browser for web automation and research using Vercel's agent-browser (https://github.com/vercel-labs/agent-browser) — navigate, click, type, extract and screenshot.

Agent Browser Skill

Give the agent a controllable headless/headed browser for tasks that need a real page: logging in, scraping JS-rendered content, filling forms, taking screenshots.

Install

npm install -g @vercel-labs/agent-browser
# or run ad-hoc
npx @vercel-labs/agent-browser

Usage pattern

1. Start the browser agent server (it exposes a local control endpoint).
2. From SSR, call it via run_command, e.g.:

   agent-browser open "https://example.com"
   agent-browser act "click the login button"
   agent-browser extract "the article title and author"
   agent-browser screenshot ./shot.png
   

When to use

Prefer web_search (Tavily) for quick factual lookups. Reach for agent-browser when you need to interact with a site — authenticated flows, dynamic content, multi-step navigation, or visual verification via screenshots.

Notes

• Honour the target site's terms of service and robots policy.
• Keep credentials out of the repo; read them from the environment.