Microsoft open-sources MXC, a policy-driven sandbox for running AI agents safely
lemire · x · 2026-10-13
Daniel Lemire highlights Microsoft's newly open-sourced MXC (Microsoft eXecution Container), already at 2.7k GitHub stars. The premise: do you trust Claude Code or Grok to run on your machine? MXC lets you write a JSON policy declaring which files a program may read/write and whether it can use the network. It's a sandboxed execution system for untrusted code (model output, plugins, tools) on Windows, Linux and macOS, with layered backends from OS-native process sandboxes to full VMs. MIT-licensed; Lemire calls it a good idea, though he notes limitations running it on a Mac.
Related event: Microsoft's Open-Source MXC Sandbox Slows git by 124x in Test(2 posts)→
More from coding & agent
- 'Spinbusy': Yacine mocks coding agents stuck in pointless fix loops — yacineMTB · 2026-10-13
- Ending agent approval fatigue: one isolated container per agent, then test the boundary — ComprehensiveShake76 · 2026-10-13
- Reverse-engineering Claude Desktop: its frontend motion tokens and CSS tricks, fully extracted — ZeYanjie · 2026-10-13
- Dev builds Little Prince 3D game in 16 days with Opus 5.5, Tripo and Blender — iamaliveix · 2026-10-13
- Indie dev renames search engine indexer to Argus, built on his own libraries — wavefnx · 2026-10-13
- Skill+ChatCut turns a product link into a 40-second 4K AI launch film — yihui_indie · 2026-10-13