Coding Without the CLI: Claude Code on the Web

Claude CodeClaude Code WebRemote ControlAnthropicGitHubCloud VMCoding AgentAutomation

Claude Code on the Web lets you run Claude Code tasks from a browser or mobile device without keeping a local terminal session open. Instead of executing inside a project folder on your machine, the web version runs the task in an Anthropic-managed cloud environment connected to a remote repository.

This is different from Claude Code Remote Control. Remote Control lets you operate a live local Claude Code session from another device. Claude Code on the Web runs the task in cloud infrastructure. Because both workflows can be reached through web or mobile surfaces, they are easy to confuse.

As of 2026-05-30, the official documentation describes Claude Code on the Web as a research preview available for Pro, Max, Team, and Enterprise users. Exact availability, resource limits, permission modes, and security behavior can change, so check the current docs before relying on this workflow for production work.

What Claude Code on the Web Is

Claude Code on the Web is a cloud-hosted version of Claude’s coding-agent workflow. It is designed to connect to a version control provider, clone a repository into a managed environment, run commands, modify files, and produce commits or pull requests.

The main value is persistence and portability. You can start a task from the browser, close your laptop, and let the cloud session continue running. This is useful for long-running refactors, test-writing tasks, documentation updates, or repository work that does not depend on your local machine.

The tradeoff is that the session does not automatically inherit local-only tools, private files, localhost services, local credentials, or custom MCP servers from your computer.

Cloud Environment

The web workflow runs each task in a cloud VM. Current docs describe a standard environment with:

  • 4 vCPUs
  • 16 GB RAM
  • 30 GB disk storage

Treat these numbers as operational details, not permanent product guarantees. Cloud execution environments can change by plan, region, product version, or Anthropic policy.

The important practical point is not the exact VM size. The important point is that this is a fresh remote execution environment. If your task depends on a local database, a local .env file, a private binary, or an MCP server running on your computer, the web session will not automatically have it.

How to Use Claude Code on the Web

Open claude.ai/code and sign in. The workflow typically requires connecting a GitHub account or installing the Claude GitHub App, then granting access to the repositories Claude should work on.

Because the web environment works from remote repositories, the code should already exist in a GitHub repository or another supported version control provider.

2. Configure the Cloud Environment

Before submitting tasks, configure the runtime environment.

Common settings include:

  • Network access: choose whether the session can access dependency registries and other network resources.
  • Environment variables: add required .env-style values for tests or build commands.
  • Setup scripts: run shell setup commands to install additional tools before the task starts.

If setup scripts are supported and cached, use them for repeatable environment preparation rather than relying on manual setup inside each session.

Select Repository, Branch, and Permission Mode

From the web interface, choose:

  • Repository
  • Target branch
  • Permission mode

Typical permission modes include:

  • Plan Mode: Claude proposes an implementation plan and waits for approval.
  • Auto Accept Edits: Claude can edit files, run commands, and push changes more autonomously.

Use Plan Mode for risky or architectural changes. Use Auto Accept Edits when the task is well-scoped, testable, and easy to review afterward.

Prompt the Task

Use task-oriented instructions that include the goal, constraints, test command, and expected output.

Example:

Write comprehensive integration tests for the auth module.
Run the test suite.
Fix any failures caused by the new tests.
Open a pull request against main with a concise summary and test results.

This works best when the repository already has scripts, tests, and conventions that Claude can follow.

Claude Code on the Web vs Remote Control

FeatureClaude Code on the WebClaude Code Remote Control
Where the code runsAnthropic-managed cloud environmentLocal machine terminal
Local file accessNo direct access to local-only filesFull access to the local project and tools
GitHub requirementUsually required because the cloud session works from remote repositoriesNot required for local directories
Session persistenceCan continue in the cloud after closing the laptopDepends on the local terminal session staying alive
Device portabilityAccessible from browser or phoneAccessible remotely, but the host machine must stay awake and connected
Local MCP accessNot available unless configured in the cloud environmentAvailable if the local session has MCP servers configured
Typical start pathOpen claude.ai/codeStart a local Claude Code session with Remote Control enabled

When to Use Claude Code on the Web

Choose Claude Code on the Web when:

  • The task can run from a clean repository checkout.
  • You want long-running work to continue without using local CPU or battery.
  • You need to start or review coding tasks from another device.
  • The repository has clear test commands and build scripts.
  • The work can be safely performed in a cloud-managed environment.

Good examples:

  • Add tests to an existing module.
  • Run a broad refactor with a clear test suite.
  • Update documentation across a repository.
  • Open a pull request after implementing a small feature.
  • Run parallel tasks against separate branches.

When to Use Remote Control

Choose Claude Code Remote Control when:

  • The code cannot leave your machine.
  • The task depends on local-only files or credentials.
  • You rely on local MCP servers.
  • You need access to localhost services, local hardware, or local development databases.
  • You want to control the same terminal session from a phone or browser.

Remote Control is effectively a remote interface for a local Claude Code session. It does not move execution to the cloud.

Practical Decision Rule

Use this simple rule:

If the task can run from a clean remote repository, use Claude Code on the Web.
If the task depends on your local machine, use Remote Control.

For sensitive repositories, private local tooling, or tasks that interact with local services, Remote Control is usually the safer fit. For long-running repository tasks that can be isolated in a cloud environment, Claude Code on the Web is often more convenient.

Best Practices

  • Keep tasks scoped and reviewable.
  • Provide the exact test, lint, or build command.
  • Use branches and pull requests rather than pushing directly to main.
  • Store secrets only through supported environment variable mechanisms.
  • Avoid granting broad network or repository access when it is not necessary.
  • Use Plan Mode for large migrations or architecture changes.
  • Review generated commits before merging.
  • Verify current resource limits and feature availability in official documentation.

Summary

Claude Code on the Web and Claude Code Remote Control solve different problems.

Claude Code on the Web moves execution to a cloud environment so coding tasks can continue without a local terminal. Remote Control keeps execution on your machine but lets you operate the session through a browser or mobile device.

The right choice depends on where the code and dependencies need to live. Use the web version for portable, repository-based automation. Use Remote Control when the task depends on your local environment.

Related