CoreWeave Code Sandbox

The Jupyter MCP Server can execute code in a CoreWeave Sandbox, a container on CoreWeave's GPU cloud. Execution is routed through the code-sandboxes coreweave engine.

tip

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

Requirements

Install the CoreWeave extra:

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

Provide credentials through environment variables:

Environment variable	Description
`CWSANDBOX_API_KEY`	CoreWeave Sandboxes API key
`CWSANDBOX_BASE_URL`	API endpoint to reach, when it is not the default `https://api.cwsandbox.com`

Configuration

Select the CoreWeave engine with SANDBOX_VARIANT=coreweave:

SANDBOX_VARIANT=coreweave
CWSANDBOX_API_KEY=your-coreweave-api-key

Or via the command line:

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

Ask for a GPU with SANDBOX_GPU, for example H100. Without it the sandbox is a plain container.

MCP client configuration:

{
  "mcpServers": {
    "jupyter": {
      "command": "uvx",
      "args": ["jupyter-mcp-server@latest"],
      "env": {
        "SANDBOX_VARIANT": "coreweave",
        "CWSANDBOX_API_KEY": "your-coreweave-api-key",
        "SANDBOX_GPU": "H100"
      }
    }
  }
}

note

State is kept by a session process the variant holds open on the sandbox's stdin, so variables set in one execution are still there in the next. Should that process fail to start, each snippet runs in a process of its own and nothing carries over. Rich display data (figures, HTML) is not returned — outputs come back as text.

Requirements​

Configuration​

Requirements

Configuration