Trust and security

What does the MCPVault security profile show?

The MCPVault security profile shows what an MCP server's code can reach, so you can decide whether that fits your setup. It opens with an exposure level, Minimal, Limited, Broad or Extensive, that describes how much of the machine and its secrets the code touches, then lists detected capabilities such as talking to external services (with the hosts named), reading environment variables in bulk, sending conversation content outward, running shell commands, loading code at runtime, reading or writing other agents' configuration folders, accessing credential files, using elevated privileges or installing itself to run later. It also says which of the core capabilities were not detected, and lists dependencies with known advisories by package, version and CVE. Each item comes from a static NVIDIA SkillSpector scan of the shipped code; documentation, tests and CI files are recorded separately and are not part of the profile. There is no score and the exposure level is not a verdict: a terminal tool is Limited because it runs commands. Read the hosts line and the detected list, and keep your client's tool approval prompts on.