Zero to Agent

English

中文
Español
Français
हिन्दी
العربية
Português
Русский

English

中文
Español
Français
हिन्दी
العربية
Português
Русский

Web and Cloud ​

📚 Series Navigation: The previous article 10 Desktop app introduced the desktop application for sandbox isolation and zero-configuration startup. This article looks further out—what if you're not at your computer, or the project is so huge it can't run locally? We'll talk about Web and Cloud usage.

"The server crashed while I was commuting, I only have an iPad on me, can I use Claude Code to urgently fix a bug?" "The company's monolithic project is 50GB, cloning it locally kills my hard drive, let alone analyzing it."

If you've encountered these two scenarios, you'll know that the local versions (CLI, IDE extensions, Desktop App) we talked about earlier have a hard limit: They must rely on your local machine's environment and performance. Once you leave your workspace, or the project size exceeds what your local machine can bear, local versions become paralyzed.

This is exactly where the Web Version (claude.ai) and Cloud integration show their strength. You don't need to configure any environment locally, you don't even need a computer, a browser is enough to connect to the codebase and let Claude start working.

After reading this article, you will get:

  • How to connect GitHub repositories on the claude.ai web version (Browser programming)
  • The Cloud Sandbox mechanism: The artifact for resolving "my computer can't run it"
  • Cloud usage scenarios: Urgent fixes on commute, architectural analysis of giant projects
  • Data privacy boundaries for cloud connections

01 Web Version Integration: The "Browser IDE" That Can Be Taken Anywhere ​

The most direct way to use Claude to write code without installing anything locally is through the claude.ai official website. But asking it to write a snippet of code and copy-pasting it yourself is too primitive; the real Web integration is letting claude.ai directly read your codebase.

Connecting to GitHub Repositories ​

claude.ai provides native integration with GitHub. The steps are very simple:

  1. Open claude.ai and log in.
  2. In the dialogue interface or settings, find GitHub Integration (usually indicated by a GitHub icon or in the integrations menu).
  3. Click to authorize, allowing Claude access to your GitHub repositories.
  4. Crucial Step: You can choose to authorize All repositories or Only specific repositories. (Strongly recommend choosing specific repositories first for safety).

After connection, what can it do?

In the chat box, you can directly mention repository files using @. Analogy: A remote "Code Reader". You say: @facebook/react/packages/react/src/React.js Explain the lifecycle here. Claude will directly go to GitHub, pull that file, read it, and explain it to you.

The biggest advantage: You didn't clone the React source code locally, you didn't configure a Node environment, you just used a browser, and you can chat with a massive open-source project. On the subway holding an iPad, you can review PRs or troubleshoot online bugs.

Limitations of the Web Version ​

Of course, the pure Web version has limitations. It is mainly a "Read-Only" or "Propose changes" state. It can give you updated code snippets, but it cannot directly modify your GitHub repository files or auto-commit (usually requires you to manually copy or use its generated diff to apply). It also cannot run npm run test for you.

💡 One-sentence summary: Web version GitHub integration = ubiquitous code reader, holding an iPad on the subway you can @ the codebase to troubleshoot bugs, but mainly read-only.

02 Cloud Sandbox: Fixing "My Computer Can't Run It" ​

If the Web version is a "Code Reader," then the Cloud Sandbox (Cloud environments like GitHub Codespaces or Replit integration) is a "Remote Heavy Machinery Plant."

Some projects are just too big. A C++ giant project takes 2 hours to compile locally; a microservice architecture requires running 8 Docker containers locally, and your 16G memory Mac starts blowing fans wildly. At this time, Claude Code running locally will also be dragged down.

The Logic of Cloud Integration ​

You can run Claude Code in a cloud environment (like GitHub Codespaces). Because Claude Code CLI is essentially a Node.js program, any cloud environment that can provide a terminal can run it.

Analogy: Hiring a worker to the remote factory. You don't pull the 50GB project back to your home garage (local); you directly rent a huge factory (Codespaces) in the cloud, then send the worker (Claude Code CLI) to that factory. He works there, compiles there, and reports the results to you.

Steps to use:

  1. Open the project in GitHub Codespaces (or other cloud IDEs).
  2. In the cloud terminal, run npm install -g @anthropic-ai/claude-code.
  3. claude to launch.
  4. Because it's running in the cloud, all file reading, writing, and executing commands happen on the cloud server.

Why is this killer?

  • Performance freedom: The cloud can allocate 32-core 64G memory, compiling takes minutes, Claude Code executes npm run build instantly.
  • Bandwidth freedom: Pulling a 10G dataset locally takes half a day; the cloud environment pulls it in seconds.
  • Environment consistency: The cloud environment is already configured, Claude Code won't get stuck on "missing local dependency".

💡 One-sentence summary: Put Claude Code CLI into Codespaces, using remote high-spec servers to do the heavy lifting, completely solving the "local compile takes 2 hours, memory blows up" dilemma.

03 Privacy and Security in the Cloud ​

When code goes to the cloud, the first thing company bosses ask is: Is the code safe?

You need to clearly distinguish two concepts:

  1. Authorization for Web Version (claude.ai): When you authorize GitHub to claude.ai, Anthropic's servers will pull your code. According to Anthropic's privacy policy, Enterprise/Team plan data is not used for model training by default, but free/Pro users need to pay attention to the terms. If it's company core confidential code, carefully evaluate before authorizing the Web version.
  2. CLI running in Cloud Sandbox (Codespaces): This is exactly the same as running locally. The code resides in your Codespaces (Microsoft's server), Claude Code CLI reads it and sends snippets to the Anthropic API. As long as your API usage does not allow data training (which standard API terms usually don't), this path is relatively more controllable.

⚠️ Safety Rule: For company private repositories, always prefer using the CLI version (whether locally or in a private cloud sandbox) over casually authorizing the Web version.

04 Summary ​

This article expanded Claude Code's capabilities beyond the local machine:

  • Web Version (claude.ai) Integration: The artifact for commuting and temporary debugging. Connect to GitHub, use @ to converse with code directly in the browser, ubiquitous but mainly read-only.
  • Cloud Sandbox (Codespaces + CLI): The ultimate solution for giant projects. Run the CLI on a high-spec remote server, leveraging cloud compute and bandwidth to crush performance bottlenecks.
  • Privacy Boundary: Distinguish the security levels between Web authorization and CLI API, protect company assets.

You should now understand: When you are away from your computer, use the Web version; when your computer can't handle the project, put the CLI in the cloud.

Next article 12 Project Initialization—We've covered all the installation and environment setups. Starting from the next article, we enter the real practical combat phase: How to use Claude Code to initialize a brand new project from zero.