In Claude Code, how does a Remote Control session differ from a cloud session in where it runs and what it can reach?
answer
- one web interface, two engines
- which machine runs the tools
- local MCP servers and files
- fresh clone in a managed VM
- which one needs the laptop awake
basics
~20 sRemote Control lets claude.ai or the Claude app drive a session that keeps running on your machine, with your files, tools and MCP servers. A cloud session runs on a fresh clone in an Anthropic-managed VM and continues after you disconnect.
solid answer
~50 sBoth show up at claude.ai/code and in the Claude app's **Code** tab, which is why people mix them up. With Remote Control (`claude remote-control`, `claude --remote-control` or `/remote-control`) the phone or browser is only a window: execution, the filesystem, local MCP servers and project configuration stay on your machine, so the computer has to stay awake and the `claude` process has to keep running. A cloud session (started at claude.ai/code, from the app, with Desktop's **Cloud** option or with `claude --cloud`) runs in an isolated VM, Anthropic-managed by default, on a fresh clone of the GitHub repository. It sees what was pushed (or uploaded as a bundle), not your live checkout, keeps working after you close the laptop and suits long, lightly steered or parallel tasks. Use Remote Control to keep steering local work in progress; use a cloud session when the task must not depend on your machine.
go deeper
Remember the one-line split: Remote Control is your own machine reached from a phone or browser; a cloud session is a fresh machine Anthropic runs for you.
Explain what follows from where the process runs: local MCP servers and uncommitted files on one side, a fresh clone, the Trusted network allowlist and survival after disconnect on the other.
Show you pick by dependency: work that needs VPN-only services or local hardware stays on Remote Control, and unattended or parallel work goes to the cloud, with the eligibility gates on Team and Enterprise.
Weigh where code executes against what leaves the network: Remote Control keeps execution local but stores the transcript remotely, while a cloud session moves the clone itself onto infrastructure you do not operate.
## Two sessions behind one interface Claude Code can be reached from a browser at claude.ai/code and from the **Code** tab of the Claude app for iOS and Android. Two very different things can sit behind that same session list: - a **Remote Control session**, which is a Claude Code process running on *your* computer that you are steering from another device; - a **cloud session**, which is a Claude Code process running on cloud infrastructure, Anthropic-managed by default or on an organization's self-hosted environment when it is routed there. The interface looks the same, but the question "which machine runs the tools?" has opposite answers, and that decides what the agent can read, run and reach. In the session list, a Remote Control session shows a computer icon with a green dot while your machine is online. ## Remote Control: a window into a local session You start Remote Control in one of three ways: `claude remote-control` (server mode, no local prompt), `claude --remote-control` / `--rc` (a normal interactive session that is also reachable remotely), or `/remote-control` / `/rc` inside a session that is already running, which carries the current conversation over. Desktop and the VS Code extension use the slash command. - **Your full local environment**: the filesystem, MCP servers, tools and project configuration all stay available, and `@` autocompletes local file paths. - **Several devices at once**: the conversation stays in sync, so you can type from the terminal, the browser and the phone interchangeably. - **Attachments**: photos go straight to Claude; other files are downloaded to your machine and passed as `@` file references. - **The cost**: the computer has to stay on and the `claude` process alive. If the laptop sleeps or the network drops, the session is offline until the machine comes back and reconnects. On a remote server you keep it alive with `tmux` or `screen`. - **Network shape**: the local process makes outbound HTTPS requests only and opens no inbound ports. While it is connected, the transcript is stored on Anthropic's servers to keep devices in sync. ## Cloud sessions: a fresh machine per task A cloud session runs in an **isolated VM** with a **fresh clone** of your GitHub repository at the branch you start from. It can be started from claude.ai/code, the mobile app, Desktop's **Cloud** environment option or `claude --cloud "task"` in a terminal. - It **keeps running** after you close the laptop, and you check on it or steer it from any device. - It sees **what was pushed**, or what Claude Code uploaded as a bundle when it cannot clone; anything installed or configured only on your machine is absent. - Its outbound network is set by the **cloud environment**; the starting `Default` environment uses the **Trusted** level, an allowlist of package registries, GitHub and cloud SDKs. - It counts against the same plan rate limits as your other Claude usage, with no separate compute charge, and it always authenticates with your subscription credentials. - Idle sessions pause and their VM can later be reclaimed; reopening one restores the conversation history but not background work that was still running. ## Side by side | | Remote Control | Cloud session | | :- | :- | :- | | Where tools execute | Your machine | Anthropic-managed VM by default | | Code it sees | Your working tree, including uncommitted and untracked files | A fresh clone; only what is pushed or bundled | | Your local MCP servers, user skills, `~/.claude` settings | Available | Not available | | Survives closing the laptop | No, it goes offline | Yes | | Typical use | Keep steering work you started at your desk | Long or parallel tasks, a repository you have not cloned | ## Choosing between them 1. If the task needs something only your machine has (a private database on the VPN, a local MCP server, a large build that needs your hardware), use **Remote Control**. 2. If the task should keep going while you are offline, or you want several independent attempts at once, use a **cloud session**. 3. If you started locally and now must leave, turn on `/remote-control`; from the CLI you cannot push a running terminal session into the cloud. Eligibility differs from a normal terminal session. Both need a claude.ai subscription login rather than an API key. Remote Control is off by default on Team and Enterprise until an Owner enables its toggle, and organizations with Zero Data Retention can use neither. Cloud sessions need Pro, Max or Team, or an Enterprise premium or Chat + Claude Code seat. ## Neighbours on the same phone Two other phone-shaped features are easy to confuse with these. **Dispatch** (Pro and Max only) lets you message a task from the Claude app that can spawn a session in the Desktop app on your machine. **Claude Code in Slack** turns an `@Claude` coding request into a cloud session under your own account. Schedules and routines that start cloud sessions on their own are a separate subject: this one is only about where a session runs once it exists.
- Can you push a session that is already running in your terminal up to the cloud?Not from the CLI: the handoff is one-way. `claude --cloud "task"` starts a *new* cloud session from your GitHub remote at the current branch, and `--teleport` pulls a cloud session down into the terminal. Only Desktop's **Open in > Cloud** moves a local Code-tab session, carrying the conversation over as a summary. To reach a running terminal session from your phone, turn on `/remote-control` instead.
- Where does a coding task you give `@Claude` in Slack actually run?In a Claude Code cloud session under your own Claude account, using your connected repositories and your plan limits; it needs cloud sessions enabled and GitHub connected at claude.ai/code. On Team and Enterprise this earlier Slack integration is being retired in favour of Claude Tag, which runs `@Claude` as an organization identity; Pro and Max keep it.
- Why can a team that routes Claude Code through Amazon Bedrock use neither feature?Both need a claude.ai account talking to Anthropic directly. Remote Control refuses on Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, or when `ANTHROPIC_BASE_URL` points at a gateway. `--cloud` refuses when a third-party provider is configured. Developers would need a separate claude.ai login, and on Team or Enterprise an Owner must also allow each feature.
- Remote Control uploads your repository and runs the session in the cloud.
- A cloud session can call MCP servers you added on your laptop with claude mcp add.
- With Remote Control you can close the laptop and the work keeps going.
- The --remote-control flag and the --cloud flag are two spellings of one feature.
- A cloud session started with --cloud keeps mirroring your checkout as you edit.