Why MCP agents keep hallucinating in big codebases (and how knowledge graphs fix it)

Last week, an agent was asked a very normal question in a very not-normal codebase: “Add audit...