browser-cloud

TestMu AI Browser Cloud — agent skill. Give any AI agent a real browser. No local Chrome. No infrastructure. Just agents that can actually use the web.

This repo contains a ready-to-use agent skill for TestMu AI Browser Cloud — a cloud browser infrastructure built for AI agents. Drop the skill into Claude, Cursor, or any LLM tool that supports custom skills/instructions, and your agent will write production-grade browser automation code on demand.

---

What This Is

The browser-cloud skill is a structured instruction file (browser-cloud/SKILL.md) that tells an LLM assistant exactly how to generate code for this domain. When you load this skill into your AI tool, it gains the ability to:

• Spin up isolated cloud browser sessions via @testmuai/browser-cloud
• Connect via Puppeteer, Playwright, or Selenium
• Handle stealth, auth persistence, tunnels, file transfer, and observability
• Wrap browser actions into LangChain, CrewAI, OpenAI function calling, or any custom agent loop

The skill is agent-agnostic — it works with any LLM or agent framework, not just Claude.

---

Install the skill

Install browser-cloud from the LambdaTest browser-cloud-skills repo with the Skills CLI:

npx skills add https://github.com/LambdaTest/browser-cloud-skills --skill browser-cloud

Requires Node.js and network access. After it finishes, your environment should pick up the skill where supported; you can still open SKILL.md manually if your tool does not integrate with the CLI.

---

Quick Start

1. Load the skill into your AI tool

Claude / Claude.ai (Projects): Copy the contents of browser-cloud/SKILL.md into a Project instruction or system prompt.

Cursor: Add to .cursor/rules/browser-cloud.mdc or paste into Cursor Settings → Rules for AI.

Any other LLM tool: Paste browser-cloud/SKILL.md as a system prompt or instruction file. The skill is plain markdown — it works anywhere.

2. Install the SDK

npm i @testmuai/browser-cloud

Node.js 16+ required. Playwright adapter requires Node.js 18+.

3. Set your credentials

# .env
LT_USERNAME=your_testmuai_username
LT_ACCESS_KEY=your_testmuai_access_key

Get credentials at testmuai.com → Settings → Account Settings.

4. Ask your agent to build something

Once the skill is loaded, ask naturally:

"Write me an agent that scrapes competitor pricing from three URLs in parallel"
"Build a LangChain tool that gives my agent access to a real browser"
"Create a browser agent that logs into our staging app and extracts the dashboard data"
"Set up a cloud browser session with stealth mode and Playwright"

The agent reads the skill, picks the right pattern, and writes complete, runnable TypeScript.

---

Repo Structure

├── package.json                              ← npm scripts and SDK dependency
├── README.md
│
└── browser-cloud/                      ← Skill bundle (copy or symlink this folder into your skills dir)
    ├── SKILL.md                              ← Load this into your AI tool
    ├── examples/                             ← Runnable TypeScript examples
    │   ├── scrape-agent.ts                   ← Batch scraping with concurrency control
    │   ├── form-fill-agent.ts                ← Form interaction with Playwright + stealth
    │   ├── auth-persist-agent.ts             ← Login once, reuse session across runs
    │   ├── parallel-research.ts              ← Parallel sessions with progress tracking
    │   └── langchain-browser-tool.ts         ← Four LangChain tool variants
    └── references/                           ← Skill reference library
        ├── patterns/
        │   ├── quick-actions.md              ← Scrape / screenshot / PDF one-liners
        │   ├── session-navigate.md           ← Multi-step navigation and interaction
        │   ├── auth-profile.md               ← Auth persistence patterns
        │   ├── parallel-sessions.md          ← Concurrent agent sessions
        │   ├── tunnel.md                     ← Localhost and internal network access
        │   └── files.md                      ← Upload/download between agent and browser
        └── integrations/
            ├── openai-functions.md         ← OpenAI + Anthropic function calling
            └── crewai.md                     ← CrewAI BaseTool (Python + Node bridge)

---

Examples

Standalone scraping agent

import { scrapeWithAgent, batchScrape } from './browser-cloud/examples/scrape-agent';

// Single URL
const result = await scrapeWithAgent('https://example.com');

// Multiple URLs, 3 concurrent sessions
const results = await batchScrape(['https://a.com', 'https://b.com', 'https://c.com'], 3);

Auth persistence across cron runs

import { fetchDashboard } from './browser-cloud/examples/auth-persist-agent';

// First run: logs in, saves session to .profiles/myapp-login.json
// All future runs: loads saved session, skips login entirely
const data = await fetchDashboard('https://app.example.com/dashboard');

LangChain agent with browser access

import { buildBrowserAgent } from './browser-cloud/examples/langchain-browser-tool';

const agent = await buildBrowserAgent();
const result = await agent.invoke({
  input: 'Research the latest pricing for our top three competitors.',
});

Parallel research across URLs

import { parallelResearch } from './browser-cloud/examples/parallel-research';

const results = await parallelResearch([
  'https://competitor-a.com/pricing',
  'https://competitor-b.com/pricing',
  'https://competitor-c.com/pricing',
], 3);

---

Supported Frameworks

Framework	Language	Status	Reference
LangChain	TypeScript	Ready	browser-cloud/examples/langchain-browser-tool.ts
OpenAI function calling	TypeScript	Ready	browser-cloud/references/integrations/openai-functions.md
Anthropic tool use	TypeScript	Ready	browser-cloud/references/integrations/openai-functions.md
CrewAI	Python	Ready (subprocess bridge)	browser-cloud/references/integrations/crewai.md
AutoGen	Python	Use OpenAI pattern	browser-cloud/references/integrations/openai-functions.md
Custom agent loop	Any	Drop-in function	Any file in browser-cloud/examples/

---

Supported Capabilities

Capability	What it does
Stealth mode	15+ fingerprint patches (Puppeteer) / 6 init scripts (Playwright). Humanized click/type timing. Randomized UA and viewport.
Auth persistence	Save login state to disk. Reload on every run. Zero re-logins for scheduled agents.
Context transfer	Pass cookies and storage between sessions in the same script run.
Tunnel	Access localhost, staging, or VPN-gated URLs from cloud browsers.
Parallel sessions	Multiple isolated browsers running concurrently. Batching support.
File transfer	Upload local files to cloud browser. Download exports, receipts, reports back to disk.
Extensions	Load Chrome extensions into sessions from S3 URLs.
Quick Actions	One-liner scrape, screenshot, and PDF — no session management required.
Observability	Every session gets video recording, console logs, and network capture in your TestMu AI automation session history.

---

Requirements

• Node.js 16+ (18+ for Playwright adapter)
• A TestMu AI account (free tier available)
• @testmuai/browser-cloud npm package

---

Contributing

Contributions welcome. The most useful additions are:

• New pattern files in browser-cloud/references/patterns/ for use cases not currently covered
• New integration files in browser-cloud/references/integrations/ for frameworks not yet supported
• New runnable examples in browser-cloud/examples/
• Improvements to browser-cloud/SKILL.md that make the LLM output more correct or more concise

Keep browser-cloud/SKILL.md under 500 lines. Every addition should remove something or replace something weaker. The skill is a decision engine, not a tutorial.

---

License

MIT