Microsoft Execution Containers (MXC) is Microsoft's open-source tool for running AI agents, or the code they create, in a fenced-off space that only reaches what it is allowed to.
In plain words
MXC treats code made by an AI model, and an agent's tools and plugins, as untrusted. A developer lists what a job needs, such as certain files and network addresses, in a JSON settings file or through its SDK, and MXC enforces that limit with a suitable container, around one tool or the whole agent. It builds on each system's own sandbox, such as Windows' AppContainer, macOS's Seatbelt or Linux's Bubblewrap. It is generally available, with its code on GitHub under the MIT licence.
A quick example
Imagine a coding agent asked to fix a website. Its policy lets it change the project folder and look at the server settings. If it tries to edit those settings, MXC stops it, whatever the agent decides.
Where you'll meet it
- Coding agents such as GitHub Copilot, OpenAI Codex and Replit.
- Running AI-written code on Windows, macOS or Linux.
Related
Sources
- DocsMicrosoft Execution Containers: policy-driven containment for AI agents, Microsoft (Windows Developer Blog) · read 11 Oct 2026
- Docsmicrosoft/mxc, Microsoft (GitHub) · read 11 Oct 2026
Checked by WhatIsAI against the sources above · 11 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.