How to run Claude Code in the cloud

You can run Claude Code in the cloud with Anthropic's Claude Code on the web, or with a cloud workspace such as Vespin, which runs Claude Code on your own Claude plan in a sandbox per agent and opens a pull request.

We build Vespin, so we have a stake in this. Most of this page is about the other ways to do it, though, and for a lot of people one of them is the better choice. The reason to leave your laptop at all is that a cloud session keeps going after you close the lid. Each one also gets its own machine and its own copy of the repository, so you can run several without them editing the same checkout, and each pushes its work to a branch you review as a pull request.

Claude Code on the web

Claude Code on the web is Anthropic's own, and if you already pay for a Claude plan it's the shortest way in. Each session runs in an isolated VM that Anthropic manages, and the VM costs nothing on top of your plan.

1. Open claude.ai/code and connect GitHub: authorize the Claude GitHub App, or run /web-setup in your terminal to use your gh login.
2. Keep the Default environment, or set network access, environment variables, and a setup script in a cloud environment.
3. Pick a repository and describe the task. From a terminal, claude --cloud "<task>" starts a session on your current branch as GitHub has it, not your local checkout, so push your commits first. That holds when the Claude GitHub App is installed on the repository. Otherwise, even with /web-setup, Claude Code uploads your local repository instead: its commits and changes to tracked files, but no untracked files. On native Windows that includes uncommitted changes to files like .env, so stash any you don't want to send.
4. Review the diff in the session and create a pull request. With Auto-fix on, Claude responds to failing checks and review comments.
5. Run claude --teleport to carry on with the session in your terminal.

Cloning and pull requests need GitHub, so a repository hosted somewhere else is harder going. A session that sits idle also loses its VM. When you reopen it you get the conversation back, but on a fresh machine (limitations). If you run one or two sessions at a time and you like working in claude.ai, this is what we'd tell a friend to use.

GitHub Actions

The Claude Code GitHub Action runs Claude Code inside your repository's workflows. You mention @claude in a comment on an issue or pull request, and it answers in that thread, pushes commits, or opens a pull request.

1. Run /install-github-app in Claude Code, or install the Claude GitHub App yourself.
2. Add a repository secret: ANTHROPIC_API_KEY for an API key, or CLAUDE_CODE_OAUTH_TOKEN for your Claude plan, made with claude setup-token.
3. /install-github-app opens a pull request with the workflow, which you merge. By hand, copy the example workflow to .github/workflows/claude.yml. It reads ANTHROPIC_API_KEY, so for a plan token change that line to claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}.
4. Mention @claude in a comment.

Each run spends GitHub Actions minutes on top of tokens. Because every exchange happens in comments, we'd reach for it for small, well-described jobs that start from an issue, and less for work that needs a lot of back-and-forth.

On a server you run

If you want the machine to be yours, any Linux VM will run Claude Code.

1. Install Claude Code on the VM and clone your repository.
2. Sign in with /login. On a machine with no browser, press c to copy the login link and open it on another computer.
3. Start claude inside tmux so it keeps running after you disconnect.
4. Run /remote-control to follow and steer the session from your phone or browser. It needs the claude.ai login from step 2, and refuses an API key or a claude setup-token token. On Team and Enterprise plans, an Owner has to turn it on first (Remote Control).

The cost of that control is upkeep. You patch and secure the server yourself, and if you want two sessions going at once, you set up a second checkout so they don't edit the same files.

The four side by side

	Claude Code on the web	GitHub Actions	Your own server	Vespin
Runs on	Anthropic's VMs	GitHub's runners	Your machine	A Vercel Sandbox per agent, or your own Mac
Pays with	Your Claude plan	API key or Claude plan	API key or Claude plan	Your Claude plan or API key
Starts from	Browser, Claude app, terminal	An @claude comment or workflow	SSH	Mac app, web app
Pull requests	Yes	Yes	You set it up	Yes
Several at once	Yes	Yes	You set it up	Yes

How we run it in Vespin

Vespin is what we built for running many coding agents at once. Each cloud agent runs Claude Code on your Claude plan in a sandbox of its own, with its own checkout of your repository. It works on a branch, opens a pull request, and fixes the checks that fail on it. When it has a question or wants a review, it posts to your inbox, so you don't have to keep opening sessions to find the one that's waiting on you.

1. Download the Mac app or open vespin.dev/app, sign in, and connect GitHub in Settings › Workspace.
2. In Settings › Agents, add your Claude plan: run claude setup-token in a terminal and paste the token it prints. You can also add an Anthropic API key.
3. Make a task, describe the work, and start an agent on it.
4. Answer its questions in your inbox and review its pull request.

Agents can also run on your own Mac, or on another machine you add, if you'd rather keep the work there. And if you only ever run one session at a time, Claude Code on the web asks less of you, since there's no token to paste.