CodeSandbox FAQs

Question 1

How can I integrate CodeSandbox MCP Server with my AI agent workflow?

Accepted Answer

Integration is typically done by running the `mcp-server-codesandbox` via `npx` in your MCP client configuration (e.g., a `mcp.json` file). You'll need to provide your `CODESANDBOX_API_TOKEN` for authentication.

Question 2

Is the CodeSandbox MCP Server stateful or stateless?

Accepted Answer

All tools and operations are stateless. Each SDK call is independent and directly executed using the provided arguments, meaning no server-managed session registry is maintained, ensuring predictable and robust interactions.

Question 3

What is the CodeSandbox MCP Server?

Accepted Answer

The CodeSandbox MCP Server exposes CodeSandbox SDK operations as tools, allowing AI agents to programmatically interact with and manage CodeSandbox development environments through a Model Context Protocol (MCP).

Question 4

What types of CodeSandbox operations can AI agents perform?

Accepted Answer

AI agents can programmatically create and configure sandboxes from templates, manage VM lifecycles (resume, hibernate), conduct comprehensive file system operations (read, write, rename, list), and manage sandbox sessions with custom permissions and environment variables.

Question 5

How does this tool ensure secure and controlled access to sandboxes?

Accepted Answer

It enables the creation of sandbox sessions with distinct 'read' or 'write' permissions. Additionally, the server can be run with a `--read-only` flag, defaulting sessions to read permission and disallowing mutating tools for enhanced security.

CodeSandbox

About

Key Features

Use Cases