> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voyager.so/llms.txt
> Use this file to discover all available pages before exploring further.

# Cloud machines over MCP

> Add the Voyager Cloud GPUs MCP server to Claude Code, Codex or Claude.ai so your agent can lease GPUs, run commands, move files and release machines as tools.

The **Voyager Cloud GPUs** MCP server gives an agent the same things as the CLI and API, as tools: lease a machine, run commands, upload and download files, open ports, pause and release, plus a Vork guide. It signs in with your Voyager account; there's no API key to paste.

```text theme={"system"}
https://mcp.voyager.so/mcp/machines
```

> [!NOTE]
>
> **Before you start:** a paid Voyager workspace (Starter or above). Machines bill by the minute until they're released. This is a separate server from the main Voyager MCP server (`https://mcp.voyager.so/mcp`, see [Connect your AI](https://docs.voyager.so/connect/overview.md)); add both if you want both.

> [tab] Claude Code
>
> ```sh theme={"system"}
> claude mcp add --transport http voyager-cloud-gpus https://mcp.voyager.so/mcp/machines
> ```
>
> Start Claude Code, type `/mcp`, choose **voyager-cloud-gpus** and sign in with your Voyager account in the browser window that opens.

> [tab] Codex
>
> ```sh theme={"system"}
> codex mcp add voyager-cloud-gpus --url https://mcp.voyager.so/mcp/machines
> ```
>
> Or add it to `~/.codex/config.toml`:
>
> ```toml theme={"system"}
> [mcp_servers.voyager-cloud-gpus]
> url = "https://mcp.voyager.so/mcp/machines"
> ```
>
> Codex opens your browser to sign in the first time it uses the server (or run `codex mcp login voyager-cloud-gpus`).

> [tab] Claude.ai
>
> 1. Open [**Settings → Connectors**](https://claude.ai/settings/connectors) and choose **Add custom connector**.
> 2. Name it **Voyager Cloud GPUs** and paste `https://mcp.voyager.so/mcp/machines`.
> 3. Click **Connect** and sign in with your Voyager account. Choose the workspace whose plan pays.
> 4. In a chat, open **+ → Connectors** and turn it on.

## Tools

| Tool | What it does |
| - | - |
| `list_presets` | Presets, GPUs and their prices |
| `create_machine` | Lease a machine |
| `get_machine` | One machine, optionally waiting until it's ready |
| `list_machines` | Your machines |
| `run_command` | Run a command, optionally waiting for its output |
| `get_job` / `job_logs` / `cancel_job` | Follow, read and stop a long command |
| `upload_files` | Send files: inline content, or a URL for the machine to fetch |
| `download_files` | Get files back: inline when small, links when large |
| `open_port` | Open a port at a `run.voyager.computer` link anyone with it can reach |
| `pause_machine` / `resume_machine` / `release_machine` | Stop billing for now, come back, or finish |
| `snapshot_machine` | Save the machine's disk |
| `vork_guide` | The guide to rendering with [Vork](https://docs.voyager.so/api/machines/vork.md) |

## Check it worked

Ask your agent: *List the Voyager cloud GPU presets and their prices.* It should call `list_presets` and show `render`, `render-lite` and `compute`.

Ask it to release machines when it's done. If it forgets, idle machines pause themselves after 15 minutes; release them from the CLI with `voyager-cloud list` and `voyager-cloud release <machine>`.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.