microsandbox
opensource self-hosted sandboxes for ai agents
claude mcp add --transport stdio zerocore-ai-microsandbox msb server start --dev \ --env MSB_LOG="info" \ --env MSB_DEV_MODE="true"
How to use
Microsandbox provides a fast, hardware-isolated environment to run untrusted workloads. The server is driven via the msb command-line tool, which lets you start a development server, pull pre-built environment images, and run sandboxed tasks. Once the Microsandbox server is running, you can access its sandbox environments (e.g., microsandbox/python) and run code inside secure sandboxes with near-instant startup times and strong isolation. Use the msb commands to create temporary sandboxes for one-off tasks or to install and manage system-wide sandboxes for frequent use. The MCP integration means you can orchestrate tasks, pass inputs, and capture outputs from within your agent workflows, enabling secure execution as part of larger pipelines or AI-assisted processes.
How to install
Prerequisites:
- curl and a shell environment
- Internet access to fetch Microsandbox installation script
Installation steps:
- Download and install Microsandbox:
curl -sSL https://get.microsandbox.dev | sh
This script installs the msb CLI and prepares your environment.
- Start the Microsandbox server (development mode):
msb server start --dev
This launches the local server in development mode for testing and iteration.
- (Optional) Pull a sandbox image to accelerate startup:
msb pull microsandbox/python
This preloads the Python sandbox image so subsequent runs start instantly.
Notes:
- The exact commands may vary slightly depending on your platform; on some systems you may need to prepend sudo.
- After installation, ensure msb is in your PATH to use the commands above.
Additional notes
Tips and common issues:
- The Microsandbox project is experimental; expect breaking changes in newer releases.
- Use msb server start --dev for a quick local server during development. For production, consult the docs for deployment guidance.
- When pulling images, you can use aliases (e.g., msb pull microsandbox/python) and then invoke sandboxes with the short alias (e.g., python).
- Environment variables in MCP configurations can be used to toggle modes or pass configuration details to the sandbox runtime. Adjust MSB_DEV_MODE and MSB_LOG as needed for debugging.
- If you encounter PATH issues, restart your terminal or source your profile to pick up the new msb executable path.
Related MCP Servers
amazon-q-developer-cli
✨ Agentic chat experience in your terminal. Build applications using natural language.
mcpcan
MCPCAN is a centralized management platform for MCP services. It deploys each MCP service using a container deployment method. The platform supports container monitoring and MCP service token verification, solving security risks and enabling rapid deployment of MCP services. It uses SSE, STDIO, and STREAMABLEHTTP access protocols to deploy MCP。
mcp-client-for-ollama
A text-based user interface (TUI) client for interacting with MCP servers using Ollama. Features include agent mode, multi-server, model switching, streaming responses, tool management, human-in-the-loop, thinking mode, model params config, MCP prompts, custom system prompt and saved preferences. Built for developers working with local LLMs.
azure-ai-travel-agents
A robust enterprise application sample (deployed on ACA) that leverages MCP and multiple AI agents orchestrated by Langchain.js, Llamaindex.TS and Microsoft Agent Framework.
mcp -code-execution-mode
An MCP server that executes Python code in isolated rootless containers with optional MCP server proxying. Implementation of Anthropic's and Cloudflare's ideas for reducing MCP tool definitions context bloat.
AutoDocs
We handle what engineers and IDEs won't: generating and maintaining technical documentation for your codebase, while also providing search with dependency-aware context to help your AI tools understand your codebase and its conventions.
