AI agents - Microsandbox

AI agents are a natural fit for microsandbox. They run tools, inspect files, install packages, call APIs, and execute generated code, often with more ambient access than they should have.

microsandbox gives those actions a dedicated microVM instead of your host process. Agents can still work normally, but their filesystem, network, lifecycle, and secrets are controlled by the sandbox boundary.

The SDK and CLI give you full programmatic control, and there are two additional ways to connect agents directly: Agent Skills for coding assistants and an MCP server for any MCP-compatible client.

Agent Skills

Agent Skills teach AI coding agents how to use microsandbox without any custom integration code. Once installed, the agent can create sandboxes, run commands, manage files, and control the full sandbox lifecycle through natural language.

Works with Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot, and other agents that support the skills format.

bash

npx skills add superradcompany/skills

This installs a set of skill files into your project that the agent reads as context. No server process, no configuration, just files that describe how to use microsandbox.

MCP Server

The microsandbox MCP server exposes microsandbox as a set of structured tool calls over the Model Context Protocol. Any MCP-compatible client can use it to manage sandbox lifecycle, execute commands, access the filesystem, work with volumes, and monitor sandbox state.

bash

claude mcp add --transport stdio microsandbox -- npx -y microsandbox-mcp

json

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "microsandbox": {
      "command": "npx",
      "args": ["-y", "microsandbox-mcp"]
    }
  }
}

json

// .vscode/mcp.json
{
  "servers": {
    "microsandbox": {
      "command": "npx",
      "args": ["-y", "microsandbox-mcp"]
    }
  }
}

json

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "microsandbox": {
      "command": "npx",
      "args": ["-y", "microsandbox-mcp"]
    }
  }
}

json

// ~/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "microsandbox": {
      "command": "npx",
      "args": ["-y", "microsandbox-mcp"]
    }
  }
}

json

// .opencode.json
{
  "mcpServers": {
    "microsandbox": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "microsandbox-mcp"]
    }
  }
}

json

// Zed settings
{
  "context_servers": {
    "microsandbox": {
      "command": {
        "path": "npx",
        "args": ["-y", "microsandbox-mcp"]
      }
    }
  }
}

</CodeGroup>

Once connected, the agent gets access to tools for creating sandboxes, running commands inside them, reading and writing files, and managing the sandbox lifecycle, all through the MCP protocol.