Daytona Code Sandbox

The Jupyter MCP Server can execute code in a Daytona cloud sandbox, with GPU and preemptible ("spot") GPU capacity available. Execution is routed through the code-sandboxes daytona engine.

tip

For the full, engine-level credential and parameter reference, see the Daytona sandbox guide in the code-sandboxes documentation.

Requirements

Install the Daytona extra:

pip install "jupyter-mcp-server[daytona]"

Provide credentials through environment variables. An API key is enough on its own; a JWT token is read together with the organization it belongs to:

Environment variable	Description
`DAYTONA_API_KEY`	Daytona API key
`DAYTONA_JWT_TOKEN`	Daytona JWT token, used instead of an API key
`DAYTONA_ORGANIZATION_ID`	Organization the JWT token belongs to, required with it

Configuration

Select the Daytona engine with SANDBOX_VARIANT=daytona:

SANDBOX_VARIANT=daytona
DAYTONA_API_KEY=your-daytona-api-key

Or via the command line:

jupyter mcp start \
  --transport streamable-http \
  --sandbox-variant daytona \
  --port 4040

Ask for a GPU with SANDBOX_GPU — for example H100, H200 or RTX-4090. Naming several, comma-separated, takes the first Daytona can find:

SANDBOX_GPU=H100,H200

Daytona also sells preemptible ("spot") GPU capacity, much cheaper than on-demand at the price of being reclaimed at any moment. It is chosen through the code-sandboxes API rather than through a SANDBOX_* variable.

MCP client configuration:

{
  "mcpServers": {
    "jupyter": {
      "command": "uvx",
      "args": ["jupyter-mcp-server@latest"],
      "env": {
        "SANDBOX_VARIANT": "daytona",
        "DAYTONA_API_KEY": "your-daytona-api-key"
      }
    }
  }
}

note

Daytona's code interpreter holds a namespace, so variables, imports and definitions from one execution are still there in the next. Rich display data (figures, HTML) is not returned — outputs come back as text.

Requirements​

Configuration​

Requirements

Configuration