Show HN: We ran 74 popular MCP servers in microVMs to see what breaks

The developers of Throne ran 74 popular MCP (Model Context Protocol) servers in isolated microVMs to test reliability and compatibility, documenting what breaks and why. The project reveals common failure points and provides a registry of tested servers with their status.

Background

MCP (Model Context Protocol) is an open standard that lets AI coding assistants (like Claude, Cursor, Windsurf, etc.) reach outside their sandbox to talk to real tools — file systems, databases, web APIs, version control. These "MCP servers" are small adapter programs that translate between the AI and the tool. A new usability study ran 74 popular MCP servers, each inside its own stripped-down virtual machine (microVM), to systematically find what breaks — missing dependencies, broken APIs, permission issues, platform-specific bugs. Key takeaway: most MCP servers have fragile setups, so tool-makers are now racing to build "sandboxed runners" that catch those failures before they reach the user. The registry at usethrone.dev collects these test results as a practical guide for developers choosing MCP servers to trust.