Gnosis turns your local documentation into a searchable knowledge base that Claude can query instead of forcing you to paste entire files into context. It indexes markdown, notebooks, commit history, and crawled websites into SQLite (or PostgreSQL at scale), then exposes hybrid BM25 + semantic search through MCP. The server ships with local ONNX embeddings so nothing hits external APIs. You get 5-10x token savings per lookup because search_docs returns ranked snippets instead of full files. It measures 92% Hit@5 on real dev docs and includes a built-in eval harness you can run against your own corpus. Useful when you're working with large codebases or documentation sets and want grounded answers without burning context on irrelevant files.
claude mcp add --transport stdio io.github.nicholasglazer-gnosis uvx gnosis