michailbul/browser-use-cloud

Browser-Use Cloud: AI Browser Automation

Purpose: Browse authenticated websites, extract content from dynamic pages, scroll carousels, fill forms, and perform computer-use tasks using cloud stealth browsers.

Use AFTER supadata fails for video/social content. Use FIRST for authenticated multi-step web tasks.

---

What This Skill Does

Runs a real Chromium browser in the cloud with:

• Anti-fingerprinting + CAPTCHA solving
• Cloudflare bypass
• Residential proxies (195+ countries)
• Persistent login profiles (reuse your sessions)

---

Trigger Conditions

Use when:

• Supadata fails to extract transcript/metadata from social video
• Extracting from an authenticated site (Instagram feed, LinkedIn, X)
• Scrolling through carousels, infinite scroll, paginated content
• Filling forms or clicking through multi-step flows
• Taking screenshots of live authenticated pages

Do NOT use for:

• Simple public URL extraction (use web_fetch)
• Video transcript (try supadata first)
• Web search

---

Pre-Flight Checks

# API key
echo "${BROWSER_USE_API_KEY:+OK}" || echo "MISSING"

# Profile (for authenticated sites)
echo "${BROWSER_USE_PROFILE_ID:+OK}" || echo "MISSING"

# SDK
npm list -g browser-use-sdk 2>/dev/null | grep browser-use-sdk || echo "NOT INSTALLED"

Auth: BROWSER_USE_API_KEY Base URL: https://api.browser-use.com

---

SDK Quick Reference (TypeScript)

import { BrowserUse } from "browser-use-sdk/v3";
import { z } from "zod";

const client = new BrowserUse(); // uses BROWSER_USE_API_KEY env var

const result = await client.run(
  "Go to <URL>, scroll through all carousel slides. " +
  "VISUALLY READ any text shown in each image and transcribe it exactly. " +
  "Extract image URLs, caption, likes, and hashtags.",
  {
    schema: z.object({ /* your schema */ }),
    model: "bu-mini",       // bu-mini ($0.002/step) or bu-max ($0.03/step)
    proxyCountryCode: "us",
    maxCostUsd: 0.50,
    profileId: process.env.BROWSER_USE_PROFILE_ID,
  }
);

---

Task Prompting Rules

1. Tell it to USE VISION — "VISUALLY READ any text shown in the image and transcribe it exactly"
2. Be specific about what's valuable — don't say "extract everything"
3. Tell it how to navigate — "Scroll through EVERY carousel slide one by one"
4. Default to bu-mini — only use bu-max if bu-mini fails

---

Fallback Chain (for video/social content)

supadata → browser-use-cloud → manual

Always try supadata first. Only use this skill when Supadata fails or when authenticated browsing is needed.

---

Models

| Model | Cost/step | Use for | |---|---|---| | bu-mini | $0.002 | Simple extraction, navigation | | bu-max | $0.03 | Complex visual tasks, multi-step reasoning |

---

Common Pattern: Instagram Carousel

const result = await client.run(
  `Go to ${url}. Scroll through EVERY carousel slide. ` +
  `For EACH slide: VISUALLY READ any text shown in the image. ` +
  `Transcribe the EXACT text. Extract image URL, caption, engagement, hashtags.`,
  {
    schema: CarouselSchema,
    model: "bu-mini",
    proxyCountryCode: "us",
    maxCostUsd: 0.50,
    profileId: process.env.BROWSER_USE_PROFILE_ID,
  }
);