Your knowledge base knows more than you think — but it also pretends to know things it doesn't.

Every time you reference a concept in a note (mentioning "reciprocal rank fusion", "the efficient frontier", "working memory decay") without actually writing a note explaining it, you're creating a phantom concept. Your AI assistant later searches for "how does reciprocal rank fusion work?" and finds a dozen passing mentions but no actual answer. Your own future self does the same.

mcp-knowledge-gaps is the audit for this. Point it at any markdown knowledge base — an Obsidian vault, Logseq graph, plain folder of .md files — and it finds every concept that appears in multiple notes but has no dedicated note explaining it. Then it ranks those gaps by how much they matter, and generates prioritized research questions to close them.