How to run Claude Code in the cloud: every option compared

Claude Code is built for your terminal, which is the problem. Your machine has to stay on, connected, and attended. If you want an agent working while you're at dinner, on another project, or asleep, Claude Code has to run somewhere that isn't your laptop.

There are three real ways to do that. Here is each one, including when ours is the wrong choice.

Option 1: your own VPS

Rent a box from Hetzner, DigitalOcean, or AWS, SSH in, and set it up.

npm install -g @anthropic-ai/claude-code
claude   # log in with your Claude account, or set ANTHROPIC_API_KEY

From there claude works as it does locally, and claude -p "your task" runs headless. Add tmux so sessions survive disconnects and you have a persistent agent box you fully control.

Good: total control, the cheapest raw compute, no third party in the loop.

Bad: you are now the platform team. Git credentials, branch hygiene, environment setup per project, and reviewing the agent's work by reading diffs over SSH. And the part everyone underestimates: the agent starts every run knowing nothing about your project except what you paste into the prompt.

Option 2: a hosted workspace

Tools like Conductor give you managed cloud workspaces: isolated machines with your repo cloned, running Claude Code or other agent CLIs in parallel. You get a real UI, isolation between runs, and pull requests instead of raw SSH.

Good: no infrastructure work, parallel agents, a proper review flow.

Bad: the agent still starts from a repo and a prompt. What you're building, what's already been decided, and what "done" means for this task live in your head or in a doc the agent never sees. A vague prompt in, a confident wrong answer out, now at cloud scale.

Option 3: a run driven by the plan

This is the gap Scope Architect fills. Instead of pointing an agent at a repo and typing a prompt, you queue a task from the project's plan, and the run starts with the full picture.

1. Scope the project once. Modules, features, and tasks with acceptance criteria, generated from an idea or the repo and refined by you.
2. Connect GitHub and your agent in Settings. Claude Code runs on your own account.
3. Queue a task. A fresh VM starts with your repo, the plan, your stack, prior decisions, and that task's criteria.
4. Watch it work. Status, logs, and comments stream to the board. If the agent hits a decision it can't make, it asks on the task and pauses. You reply and it resumes.
5. Review the PR. Work lands on its own branch and comes back as a pull request. Nothing touches your default branch until you merge.

A VM is a VM. The difference is what the agent knows when it starts, and where its questions go while it works.

Which one

• You want a lab and you enjoy infrastructure: the VPS. It's a good way to learn how agents behave unattended.
• You want several raw agents on one repo today: a hosted workspace does that well.
• You want the run to be right the first time and visible while it happens: the plan has to travel with the task. That's what cloud runs are for. They're on every paid plan from $30 a month, and runs bill to your own agent account or API keys with no markup.

One warning that applies to all three. Unattended agents amplify whatever you give them. A sharp, scoped task comes back as a mergeable PR. A vague one comes back as forty files of confident wrong. The cloud doesn't fix prompts. Plans do.

Next: Claude Code keeps losing context. Give it a plan instead of a paragraph.