How do you give Claude Code or Codex its own cloud computer?

Quick answer: Install the Orgo CLI, run orgo login once, and tell Claude Code or Codex to run orgo manifest first, so it learns every real command before it creates and drives an Orgo cloud computer from the terminal. For Claude Code, the Orgo skill adds one file that teaches it when to reach for that computer on its own.

TL;DR

• What you get: a full cloud desktop with a browser and a terminal, which your coding agent controls through shell commands it already knows how to run (llms.txt).
• Install: curl -fsSL https://orgo.ai/install.sh | bash on macOS or Linux (brings its own Node, no sudo), or npm install -g orgo on Node 20 or later, which is also the way on Windows (CLI guide).
• Sign in: orgo login runs a device-code flow in your browser and saves a credential to ~/.orgo/credentials.json. In CI, set ORGO_API_KEY instead.
• Teach the agent: orgo manifest prints every command and option as JSON, so the agent works from real flags instead of guessing.
• Drive it: orgo computers bash, orgo computers screenshot, orgo run "<prompt>", and orgo ssh cover shell work, screen checks, whole tasks, and an interactive terminal. Every command takes --json.
• Claude Code shortcut: drop the Orgo skill into ~/.claude/skills/computer/ and Claude picks it up on its next start (Claude Code skill).

Orgo provides the computer, not the agent. Your coding agent stays the brain; Orgo is where it gets a screen, a browser, and a machine that is not your laptop.

Why would a coding agent need its own computer?

Claude Code and Codex are good at reading, editing, and running code in your repo. Some tasks need more than that: a real browser to check how a page renders, a GUI app to click through, an installer you do not want on your own machine, or a session that stays up after you close the terminal.

Orgo's own skill file puts the test plainly: does the task need vision of a rendered page or app, persistence across sessions, or isolation from the user's machine? If none of those apply, the agent should use its normal tools and skip the computer (SKILL.md).

An Orgo computer comes online in seconds and runs until you stop it (llms.txt), or, if it runs by the hour, until its credit runs out. Linux and Windows are self-serve. Macs and iPhones come from stock through Get one in the app, while any are ready (pricing). New to the idea of an agent driving a desktop? Start with What is computer use? and Cloud desktops vs. headless browsers.

How do you set it up in five steps?

1. Install the CLI.

   bashCopy

   curl -fsSL https://orgo.ai/install.sh | bash
   # or, with Node 20+ (and on Windows)
   npm install -g orgo

   The script puts an orgo wrapper in ~/.local/bin, so make sure that folder is on your PATH.

2. Sign in once. orgo login opens your browser and completes a device-code flow. On an SSH session or a headless box, orgo login --no-open prints the URL instead of opening a browser. The credential is saved under ~/.orgo/credentials.json with mode 0600 (CLI guide).

3. Pick a workspace. orgo workspaces use "my-workspace", then orgo computers list to see what is already running.

4. Give the agent the command list. Paste a prompt like this into Claude Code or Codex:

   Plain textCopy

   I want you to use the Orgo CLI. First run `orgo manifest` to learn every
   command and flag, then run `orgo computers list` and tell me what you see.

5. Let it create and use a computer.

   bashCopy

   orgo computers create --workspace my-workspace --name agent-1 --ram 8 --cpu 2
   orgo computers bash agent-1 -c "uname -a"
   orgo computers screenshot agent-1 -o shot.png

   8 GB and 2 vCPU is a Medium, the one size that runs by the hour with no plan. Other sizes need a plan. With no plan, a create that names another size still makes a Medium instead of failing, so read ram and cpu in the --json answer.

When stdout is not a terminal, the CLI switches to plain text with no prompts, so an agent calling it from a tool gets clean output. Pass --json or set ORGO_JSON=1 for machine-readable results (CLI guide).

Which CLI commands matter most to a coding agent?

Source: CLI guide and orgo <command> --help.

What the create flags accept:

• --ram takes 4, 8, 16, 32, or 64 GB. --cpu takes 0.5, 1, 2, 4, or 8.
• One computer tops out at 8 vCPU, so on a plan --cpu 16 is refused with 403 PER_COMPUTER_CPU_CAP.
• On a plan, 8 vCPU or 64 GB answers 403 CONTACT_US_FOR_SIZE unless Orgo has enabled sizes past XL for your account (Create a computer).
• Set --os to linux (default) or windows. Macs and iPhones come from stock through Get one in the app, not from this flag.
• Windows is included on the Scale plan. On Starter or Growth you add a Windows computer for $47/mo, which is a $15 license plus a Medium's memory, vCPU and disk. Without one the create answers 403 WINDOWS_REQUIRES_SCALE (pricing).

Two more flags worth knowing: bash and exec take --timeout in seconds (up to 300), and orgo run takes --model (default claude-sonnet-5).

orgo ssh is a terminal over a WebSocket, not an SSH server. For a standard SSH endpoint you reach with your own key, see Connect with SSH.

How does the Claude Code skill work?

The Orgo skill is a single SKILL.md file for Claude Code. Install it for yourself or for one project:

# personal
mkdir -p ~/.claude/skills/computer && \
  curl -fsSL https://orgo.ai/skills/computer/SKILL.md \
    -o ~/.claude/skills/computer/SKILL.md
 
# one project
mkdir -p .claude/skills/computer && \
  curl -fsSL https://orgo.ai/skills/computer/SKILL.md \
    -o .claude/skills/computer/SKILL.md

Restart Claude Code if it is running. Claude reads the skill's metadata at startup (about 100 tokens) and loads the full body only when a request matches, such as "spin up an orgo computer and screenshot the orgo.ai homepage" (Claude Code skill).

What the skill tells Claude to do:

• Use one-shot orgo run with --json, the orgo agent TUI, orgo ssh, and orgo computers create.
• Reuse a computer that is already running before it provisions a new one, and stop computers when finished.
• Ask Orgo's support agent when it is stuck instead of guessing.
• Skip the computer for tasks its Read, Edit, and Bash tools already handle.

The skill also works on claude.ai (upload a .zip of the computer/ folder) and through the Claude API's /v1/skills endpoints. It is plain markdown, so read it before you install it, as you would any package.

CLI, MCP, or API: which should a coding agent use?

They reach the same computers, so you can mix them. The local MCP server reads the same ~/.orgo/credentials.json the CLI writes. Codex can add it with codex mcp add orgo --env ORGO_API_KEY=sk_live_... -- npx -y orgo-mcp-server (MCP). For the MCP path in detail, see Orgo MCP for AI agents. For one request that runs a whole task, see Orgo computer use API.

How do you use it in CI or a headless box?

Skip orgo login and set ORGO_API_KEY. The CLI uses that key directly instead of the credentials file (CLI guide). One catch: if ORGO_PROFILE or --profile is set, the CLI reads that saved profile and ignores ORGO_API_KEY, so leave both unset in the job.

Which key to make depends on what the job does:

A scoped key is limited to computers, files and agent runs. Template, account and stored-image routes refuse it with 403 workspace_scope_mismatch, which the CLI shows as "Template not found" or a failed screenshot download. Start with a scoped key, and switch to an account-wide one only when the job needs a template.

A run with a workspace-scoped key, passing every create flag so nothing prompts:

export ORGO_API_KEY=sk_live_...   # scoped to my-workspace
orgo computers create --workspace my-workspace --name ci-run --ram 8 --cpu 2 --json
orgo computers bash ci-run -c "uname -a"
orgo run "open chrome and check the staging homepage loads" --computer ci-run --json
orgo computers stop ci-run

If the job needs the same setup every time, use an account-wide key and launch from a template. For example, orgo computers create --workspace my-workspace --name ci-run --template coding gives a computer with Claude Code and Codex installed. See Orgo templates and triggers. To keep a computer's state between runs, see How do you pause, fork, or clone an AI agent's computer?

FAQ

Does the Orgo CLI need Node installed?

Not with the install script. It downloads its own Node into ~/.orgo/node and needs no sudo. The script supports macOS and Linux. On Windows, or anywhere you prefer npm, npm install -g orgo needs Node 20 or later (CLI guide).

Can Codex use the Orgo CLI?

Yes. The CLI is a normal shell tool, and Orgo's docs recommend telling Claude Code, Codex, or another coding agent to run orgo manifest before it uses the CLI (CLI guide). The packaged skill file is written for Claude Code. Codex can also use Orgo through MCP.

Is orgo ssh real SSH?

No. It is an interactive terminal over a WebSocket. For a standard SSH endpoint with your own key, use Connect with SSH.

Which model does orgo agent use?

The TUI defaults to Sonnet 5, and /model switches to Opus 5. Sessions save to ~/.orgo/sessions/ after every turn, and orgo resume picks one up again (CLI guide).

What does it cost?

A Medium Linux computer (2 vCPU, 8 GB RAM) runs by the hour from credit at $0.055 an hour, or $0.0027 an hour while stopped, as of November 2026. Other sizes and Windows need a plan: Starter ($29/mo), Growth ($99/mo), Scale ($399/mo), or Enterprise. Turns from orgo run and orgo agent are billed as AI usage on top of the computer. A new account can get $5 of starting credit after the free checks (pricing).

Can my coding agent sign up for Orgo without me?

Yes. Agent signup (POST https://www.orgo.ai/api/v1/agents) gives the agent its own account and API key with $2 of starting credit, with no email, card or person involved. That credit pays for computers, not AI. The account expires 72 hours after signup unless it buys credit or a person claims it with the claim link it returns (For agents, Agent signup). To put the agent on your own account instead, use orgo login: the agent shows a short code and you approve it in a browser, so no key is pasted into a chat (For agents).

Related

• Orgo MCP for AI agents
• Orgo computer use API: run an AI agent with one request
• Run Claude computer use on Orgo
• Orgo templates and triggers
• How to scale AI agent computers to thousands on Orgo