Oct 27, 2026

How to Use Claude Code in the Cloud from Your Browser or Phone (claude.ai/code Setup Guide)

You can run Claude Code without installing anything: open claude.ai/code in a browser (or the Code tab in the Claude phone app), connect GitHub once, pick a repository, type the task, and Claude works on it in a cloud computer that keeps going after you close your laptop. It is included in Pro, Max and Team plans (and Enterprise seats that include Claude Code), it uses the same usage limits as the rest of your Claude plan, and there is no extra charge for the cloud machine. This guide walks through the setup screens, which options to pick, how to review and merge what Claude produces, how to use it from your phone, and how to fix the errors people hit most often. Everything below was checked against Anthropic’s official Claude Code documentation on October 27, 2026.

Key facts (from the official cloud docs and quickstart):

Plans: Pro, Max, Team, and Enterprise users with premium seats or Chat + Claude Code seats.

Cost: counts toward your normal Claude usage limits; no separate charge for the cloud VM. Parallel sessions use your limits faster.

Needs GitHub: cloud sessions clone from GitHub and push branches there. Private repositories need the Claude GitHub App installed.

Where it runs: a fresh Ubuntu 24.04 virtual machine per session, with roughly 4 vCPUs, 16 GB of RAM and 30 GB of disk (cloud environments docs).

Keeps running: closing the tab does not stop the session. It finishes the current task and then idles.

When the cloud version is the right choice

Claude Code behaves the same everywhere; what changes is where it runs. Anthropic’s docs describe three ways to run it, and it helps to know which one you actually want before you set anything up.

	Cloud session	Local session	Local session with Remote Control
Runs on	A cloud VM managed by Anthropic	Your computer	Your computer
Start it from	claude.ai/code, the Claude phone app, the Desktop app with Cloud selected, or the terminal	Terminal, IDE, or Desktop app with Local	Terminal, VS Code extension, or Desktop app
Uses your own computer’s settings and tools	No, only what is in the repository	Yes	Yes
Needs GitHub	Yes	No	No
Keeps going if you disconnect	Yes	No	Only while your computer stays on

Source: Compare ways to run Claude Code.

Anthropic says cloud sessions work best for:

Running several tasks at once. Each task gets its own session and its own branch, so you don’t wait for one to finish before starting the next.
Repositories you don’t have on your computer. Claude clones a fresh copy every time.
Well-defined jobs you don’t need to babysit. Submit the task, do something else, review the result later.
Questions about a codebase. “How does login work in this project?” without downloading anything.

Pick Remote Control or a local session instead when the work depends on things that only exist on your computer: files that aren’t in the repository, apps you installed, company VPN access, or browser sign-ins. If what you need is an agent working inside your signed-in browser, our guide to letting AI agents use your logged-in Chrome covers that route.

Before you start: a quick checklist

A Claude plan that includes cloud sessions. Pro, Max or Team, or an Enterprise seat that includes Claude Code. On Team and Enterprise, if your organization uses custom roles, you also need a role with the cloud sessions permission; an Owner adds it under Organization settings > Roles.
A GitHub account and a repository. Cloud sessions work with existing repositories. If you’re starting from nothing, create a new repository on GitHub with a README first (GitHub’s “New repository” page has a checkbox for this).
For Team and Enterprise only: an Owner must turn on the GitHub connector at Organization settings > Connectors. Until they do, you’ll see “GitHub access is required for Claude Code cloud sessions” instead of a sign-in button.

There are also organizations that can’t use cloud sessions at all: Anthropic’s quickstart says organizations with Zero Data Retention, the HIPAA configuration, or customer-managed encryption keys can’t start them. If your company IT has turned on IP allowlisting for Claude, every Anthropic-hosted cloud session fails with an authentication error until support grants an exemption (Limitations).

Go to claude.ai/code and sign in with your normal claude.ai account.
You’ll be prompted to connect GitHub. Follow the prompt; GitHub shows an authorization page. Approve it and you’re sent back to claude.ai/code.
For private repositories, install the Claude GitHub App. The browser connection alone lets sessions clone any public repository. For private ones, install the Claude GitHub App on each GitHub account or organization that owns them. On a company GitHub organization, an organization owner may have to approve it. If onboarding offers the install and you’d rather wait, click Skip.

Installing the GitHub App also unlocks Auto-fix (covered below), which lets Claude respond to failed checks and review comments on your pull requests.

Already use the GitHub command-line tool? There is a terminal shortcut, /web-setup, that sends your existing GitHub CLI login to Claude instead of using the browser flow. Most readers can ignore it. If you do use it, know that your cloud sessions can then reach every repository that login can reach, and you can remove the connection later at claude.ai/customize/connectors.

Step 2: Accept the Default environment

An environment is a saved set of rules for the cloud computer: what websites it may reach, any settings it should have, and anything to install before Claude starts.

Pro and Max: onboarding creates one called Default for you automatically.
Team and Enterprise: you’ll see a Create your first cloud environment form. Leave the prefilled name and network access as they are and click Create & finish. (If an Owner turned on Quick setup, Default is created for you.)

Default uses Trusted network access, meaning the cloud computer can reach common package registries and a list of approved developer sites, and nothing else. For a first project, Anthropic’s advice is to use Default as it is. You’ll only need to change it later if Claude reports that it can’t reach a site or a tool it needs.

Step 3: Start your first task

On the new-session screen at claude.ai/code:

Choose a repository. Click the repository selector below the message box. You can add more than one repository to the same session.
Choose a branch (optional). The control after the repository selector lets you start from a branch other than the default.
Choose a permission mode. Cloud sessions offer three:
- Plan: Claude proposes an approach and waits for your approval before changing any files. The safest choice for your first few tasks.
- Accept edits: Claude makes changes and pushes a branch without stopping to ask.
- Auto: a safety classifier reviews Claude’s actions instead of asking you. It only appears if your organization allows it and the selected model supports it.
Cloud sessions don’t offer Manual or Bypass permissions. Our Claude Code permissions and auto mode guide explains these modes in more depth.
Describe the task and press Enter. Be specific. Anthropic’s examples: “Add a README with setup instructions” or “Fix the failing auth test in tests/test_auth.py” work better than “fix tests”. Paste any error message you have, and describe what should happen, not just what’s broken.

Claude then clones the repository, sets up the cloud computer, and starts working. You can watch and type corrections while it works, or close the tab and come back later.

Prompts that work well in a cloud session

Because you can’t see the cloud computer’s screen and you’re often not watching live, it pays to front-load your instructions:

“Read the project first and tell me in plain English how sign-up works. Don’t change anything.” (Good for Plan mode.)
“Fix the typo on the pricing page, run the tests, and push a branch. Stop and ask me if the tests were already failing before your change.”
“Update the README so a non-technical teammate could install this. Keep it under one page.”

You don’t get a terminal inside the cloud computer. Claude runs every command for you, so if you want tests run or a database started, just say so in the prompt (cloud environments docs).

Changed your mind about a message? If you send a message while Claude is working, it waits in a queue. Click the ✕ on the queued message to take it back and edit it. Once Claude has read it, it stays.

Step 4: Review the changes and open a pull request

When Claude reaches a stopping point, it pushes a branch to GitHub. The session stays open, so creating a pull request and asking for more changes all happen in the same conversation.

Open the diff. A Changes card in the panel beside the chat shows how many lines were added and removed, for example +42 -18. Click those numbers to see every change. (If the panel is hidden, use the Show session rail icon in the session’s title bar or its ⋮ menu.)
Comment on specific lines. Select a line in the diff, type feedback such as “don’t remove this check”, and press Enter. Comments pile up and are sent together with your next message, so Claude knows exactly which line you mean.
Create the pull request. Click Create PR on the Changes card and Claude writes the title and description. To open a draft PR instead, right-click Create PR, choose Create draft PR, then click the button.
Keep iterating. If a check fails or a reviewer leaves comments, paste them into the chat and ask Claude to address them.

Turn on Auto-fix so Claude watches the PR for you

With the Claude GitHub App installed, Claude can watch a pull request and react on its own when a check fails or a reviewer comments. To turn it on for a PR created in a cloud session, open the ⋮ menu next to the pull request in the Changes card and switch on Auto-fix. For any other PR, paste its link into a session and tell Claude to auto-fix it. To stop, switch Auto-fix off in the same menu or tell Claude to stop watching.

How it behaves, per Anthropic’s docs:

Clear fixes: Claude makes the change, pushes it, and explains what it did.
Ambiguous or architecturally significant comments: Claude asks you first.
Duplicate or no-action events: Claude notes them and moves on.

Two gotchas worth knowing before you turn it on:

It can’t see merge conflicts. GitHub sends no notification when the main branch moves ahead and creates a conflict, so open the session and ask Claude to rebase.
Its replies appear under your GitHub name. They’re labeled as coming from Claude Code, but if your repository has automation triggered by PR comments (Anthropic names Atlantis, Terraform Cloud and custom GitHub Actions on issue_comment), a Claude reply can trigger it. Consider leaving Auto-fix off where a comment can deploy something.

Using it from your phone

There is no separate Claude Code app. Install the regular Claude app for iOS or Android, sign in with the same account and organization, and tap Code (mobile docs). From there you can pick a repository and branch, describe a task, and submit it, or open a running session to check progress, answer Claude’s questions, or redirect it. Sessions follow you across devices: a task started on your laptop can be reviewed on your phone, and vice versa.

If you don’t see the Code tab, your plan or organization may not include it. If your computer will be switched off, use cloud sessions; Remote Control only works while your own computer is on and running Claude Code.

Your sessions are listed in the sidebar at claude.ai/code.

Sharing on Pro or Max: set the session’s visibility to Public, which lets anyone logged in to claude.ai open the link. Repository-access checks are not on by default, so read through the session first: Anthropic warns it may contain code and credentials from private repositories. You can require recipients to have repository access, or hide your name, under Settings > Claude Code > Sharing settings.
Sharing on Team or Enterprise: visibility options are Private and Team, repository-access verification is on by default, and your display name is visible to recipients.
People you share with see the latest state when they open the link, but it doesn’t update live.
Archive hides a finished session (sidebar, ⋯ menu, Archive). Archived sessions are on the Archived tab at claude.ai/code/sessions.
Delete removes a session and its data permanently and can’t be undone; you’re asked to confirm.

What carries over from your own Claude Code setup (and what doesn’t)

This trips up people who already use Claude Code on their computer. A cloud session starts from a fresh copy of the repository, so anything saved in the repository comes along and anything saved only on your computer does not (what carries over).

Comes along	Stays behind
The repository’s `CLAUDE.md` instructions file	Your personal `~/.claude/CLAUDE.md` on your computer
Skills, agents and commands saved in the repository’s `.claude` folder	Personal skills, agents and commands saved only on your computer
MCP servers listed in the repository’s `.mcp.json` (single-repository sessions)	MCP servers you added only for yourself
Your organization’s managed settings	Plugins (even ones the repository turns on), and sign-ins like AWS SSO that need a browser

The fix is simple to say to your agent: “commit our project instructions and skills into the repository so cloud sessions can use them.” Skills you enable on claude.ai itself load in cloud sessions automatically. If you write your own skills, our guide to writing Claude skills covers where to save them.

Changing network access (only if Claude gets blocked)

If Claude reports that it can’t download something or reach a website, the environment’s network setting is the likely cause. To change it, go to the new-session screen at claude.ai/code, click the first control under the message box (the environment selector), hover over your environment, click the settings icon, and open Network access. There are four levels (access levels):

Level	What the cloud computer can reach
None	Nothing through the session’s network
Trusted (default)	Package registries, GitHub, cloud SDKs and other approved developer sites
Full	Any site
Custom	Only the sites you list, one per line, optionally plus the Trusted list

A few things reach the session no matter which level you choose: GitHub (through a separate proxy), MCP connectors you turn on, and Anthropic’s own API. Anthropic notes that even with None, Claude Code can still talk to the Anthropic API, which may allow data to leave the VM. Changes to network access apply to running sessions within about a minute.

Don’t paste passwords or API keys into Environment variables. Anyone who uses the environment can read them. For a key that Claude needs to call a service, use the environment’s Network secrets row instead: Anthropic’s proxy adds the key to requests for the sites you list, so the key never sits inside the cloud computer. On Team and Enterprise, who can add secrets depends on your role.

Troubleshooting the common errors

No repositories appear after connecting GitHub. Public repositories work with the browser connection alone; private ones appear only after the Claude GitHub App is installed on the account or organization that owns them, with access to that repository. Install it or ask your organization owner to approve it.

The page only shows a GitHub login button. Cloud sessions require a connected GitHub account. If you don’t want to connect GitHub, use Remote Control to run Claude Code on your own computer and control it from the browser or phone.

“Claude Code isn’t available on your account.” Your seat in that organization doesn’t include Claude Code. Click Switch organization if you belong to another one, or ask an Owner for a seat that includes it.

“GitHub access is required for Claude Code cloud sessions.” (Team and Enterprise.) An Owner needs to turn on the GitHub connector at Organization settings > Connectors. Then reload claude.ai/code.

“Session creation failed” or stuck at provisioning. Anthropic couldn’t allocate a cloud computer. Check status.claude.com, wait a minute and retry, and confirm your GitHub connection can see the repository.

“Your new session couldn’t start.” Your prompt is put back in the box. Check your internet connection or VPN and send again; if it repeats, check the status page.

Setup script failed / sessions hang during setup. If you (or a teammate) added a setup script to the environment, it must finish without errors and in roughly five minutes. Ask your agent to shorten it, or remove the script to confirm it’s the cause.

The session kept running after I closed the tab. That’s by design. Archive or delete it from the sidebar when you’re done.

“Environment expired.” Idle sessions eventually have their cloud computer reclaimed. Reopen the session from claude.ai/code: your conversation comes back, but commands or helpers that were still running in the background don’t. Ask Claude to rerun what you still need. A session waiting for you to approve a connector or sign in counts as idle and can expire during the wait.

Errors mentioning Amazon Bedrock, Google Vertex AI or an API key. Cloud sessions only work with a claude.ai sign-in on Anthropic’s own plans. They aren’t available through third-party providers or with an API key.

Limits to plan around

Usage: cloud sessions share rate limits with all your other Claude and Claude Code use. Running five tasks in parallel uses your allowance about five times as fast.
GitHub only for pushing: cloning and pull requests require GitHub (GitHub Enterprise Server is supported on Team and Enterprise). GitLab or Bitbucket projects can be sent up as a one-off copy from the terminal, but the session can’t push results back.
Machine size: roughly 4 vCPUs, 16 GB of RAM and 30 GB of disk. Very large builds may be stopped; use Remote Control on your own hardware for those.
Long commands: Claude waits up to 2 minutes for a command by default (it can ask for up to 10), and longer jobs move to the background for up to 30 more minutes.

FAQ

Is Claude Code on the web free? It’s not a separate product with its own price. It’s included in Pro, Max and Team plans (and eligible Enterprise seats) and draws from the same usage limits; there’s no additional charge for the cloud computer.

Do I need to know how to code? No setup on your computer is required, and you can describe tasks in plain English and review changes in the diff view. You do need a GitHub account, and it helps to start in Plan mode so you approve the approach before anything changes.

Can it work on a project that isn’t on GitHub? Not fully. It can read a copy sent from the terminal, but it can’t push results back to GitLab, Bitbucket or other hosts.

Does closing my laptop stop the task? No. The session keeps running on Anthropic’s computer until the current task is done, then idles. Check back from any browser or the Claude phone app.

How is this different from Remote Control? Remote Control lets your phone or browser steer a Claude Code session that’s running on your computer, with your files and tools. Cloud sessions run on Anthropic’s computer and only see what’s in the repository.

Is my code safe? Each session runs in its own isolated cloud computer, network access is limited by default, and your GitHub credentials stay outside that computer behind a proxy (security and isolation). Be careful with public sharing on Pro and Max, which doesn’t check repository access by default.

If you want to get more out of Claude Code generally, our Claude Code daily driver guide is a good next read, or browse the latest tutorials on the 7minAI homepage.

Optional appendix: for people who also use the terminal

If you already run Claude Code on your computer, these official shortcuts connect it to cloud sessions (move tasks between terminal and cloud):

/web-setup inside Claude Code connects GitHub using your existing GitHub CLI login and creates a Default environment if you don’t have one. It requires a claude.ai sign-in (check with /status), not an API key.
claude --cloud starts a cloud session from your terminal.
/remote-env picks which environment terminal-started cloud sessions use.
/autofix-pr, run on a pull request’s branch, starts a cloud session with Auto-fix switched on.
/mobile shows a QR code that opens the right app store for the Claude phone app.

You can also pre-fill a new session by link, for example https://claude.ai/code?prompt=Fix%20the%20login%20bug&repositories=acme/webapp, which is handy for a button in an issue tracker. The supported parameters are prompt, prompt_url, repositories and environment (pre-fill sessions).