Overview
Code Sandbox MCP is a secure, Docker-based sandbox for AI apps to safely execute arbitrary code in isolated containers. It provides flexible container management using any Docker image as the execution environment, along with file transfer between host and sandbox and the ability to run shell commands inside the container. Real-time logging streams stdout and stderr, enabling live monitoring of activities. The MCP includes auto-updates for the binary and update checks, and it is multi-platform (Linux, macOS, Windows). It exposes a set of tooling commands such as sandbox_initialize (default image python:3.12-slim-bookworm), copy_project, write_file, copy_file, sandbox_exec, and sandbox_stop to manage the lifecycle of sandboxes. A Container Logs Resource at containers://{id}/logs provides access to logs as text. Security is emphasized by isolated execution environments and resource constraints. The project offers installer-based and manual installation methods, and config integration with Claude Desktop, allowing other AI apps to use code-sandbox-mcp as the execution backend.
Features
Flexible Container Management
Create and manage isolated Docker containers for code execution.
Custom Environment Support
Use any Docker image as your execution environment.
File Operations
Easy file and directory transfer between host and containers.
Command Execution
Run any shell commands within the containerized environment.
Real-time Logging
Stream container logs and command output in real-time.
Auto-Updates
Built-in update checking and automatic binary updates.
Multi-Platform
Supports Linux, macOS, and Windows.
Container Logs Resource
Access container logs via a dynamic resource path containers://{id}/logs (text/plain).
