Microsandbox, by Superradcompany, is an mcp platform designed for hardware-isolated execution environments. Built primarily with Rust, it functions as an MCP server, providing a secure backend for AI agents and developers. It supports local microVM management and can connect to a cloud service. SDKs are available for Rust, TypeScript, Python, and Go, offering broad integration capabilities for secure code execution and development workflows across multiple operating systems.

What does Microsandbox connect to?

Microsandbox manages local microVMs to provide hardware-isolated execution environments. It operates as an MCP server, serving as a secure execution backend for AI agents. It utilizes hardware virtualization technologies such as KVM on Linux, Apple Silicon's Hypervisor.framework on macOS, and Windows Hypervisor Platform (WHP) on Windows. It also supports running OCI-compatible container images within these environments. When configured with an API key, Microsandbox can connect to a "microsandbox cloud" for extended functionality.

Core features and tools

The MCP server exposes structured tool calls for diverse functionalities. These include sandbox lifecycle (creating, managing, inspecting, listing, stopping, deleting sandboxes) and command execution (executing commands, attaching shells). It provides filesystem access (reading, writing, listing, copying, stat, streaming operations) and volumes (managing named volumes, bind mounts, tmpfs mounts, disk-image mounts). Other tools cover monitoring (metrics, logs, snapshots), network rules (policies, DNS filtering, TLS interception, port publishing), secrets, image management, rootfs patching, and managing registry authentication, SSH, and SFTP.

Who is it for?

Microsandbox targets developers, AI agents, and anyone requiring secure execution of untrusted code. This includes those developing plugins, running CI jobs, setting up disposable development environments, performing web scraping, or automating tasks securely. It is particularly valuable for scenarios involving AI-generated code that requires hardware-level isolation to prevent host system compromise.

Skip this if your workflow does not involve executing untrusted code or AI-generated code in isolated environments.

How to get started

The server authenticates using an API key passed as an environment variable, such as SUPERSERVE_API_KEY, and primarily supports stdio as a transport protocol for local interactions.

Use cases

• Secure AI Agent Code Execution: AI agents safely create and manage sandboxes, execute generated code, install dependencies, and retrieve results without compromising the host system.
• Running Untrusted User Code/Plugins: Execute user-submitted code or third-party plugins in isolated microVMs to prevent potential malicious actions from affecting the host.
• CI/CD and Disposable Development Environments: Run continuous integration jobs or create ephemeral development environments within fresh, isolated microVMs, ensuring consistency and security.

Microsandbox provides a secure and isolated execution environment for diverse applications, from AI agents to CI/CD pipelines. Its reliance on hardware virtualization and a daemon-less local runtime offers strong security and ease of use for untrusted code execution. The platform's modular toolset enables fine-grained control over sandboxes, filesystems, and networking.