Documentation

How It Works

The architecture, why it is stateless, how responses are shaped, and the security model, for anyone who has to approve running it.

You do not need any of this to use the server. It is here because someone in your organisation will eventually ask what it does with your data, why it can be trusted, and what happens when it is wrong. “It’s an MCP server” is not an answer.

The short version

Nothing is stored. No sessions, no per-user state, no database. Protocol revision 2026-07-28 removed sessions from MCP entirely, so a server that holds nothing between requests is now the protocol’s native shape rather than a compromise.

Nothing is executed. No sandbox, no code execution, no repository access. The server reads a vulnerability database and returns shaped answers.

No code is sent to Vulnetix. There is no tool that accepts file content, because there is nothing on the other side that could use it. What crosses the wire is a package name, a version, or an advisory id.

Your credential is forwarded, never interpreted. The server does not hold the secret needed to validate it and does not try. The API is the authority; the server is a relay.

The AI layer is optional and cannot break a tool. Every AI-augmented tool returns its complete database result regardless of whether inference ran.