obsidian-mcp-semantic-metagraph

obsidian-mcp-semantic-metagraph is a fork of lstpsche/obsidian-mcp that provides an MCP server for AI agents to use semantic search, lexical search, and graph navigation over an Obsidian vault. It reads the vault directly from disk, without an Obsidian plugin or REST API. The fork rebuilds the semantic retrieval layer: it indexes the note body as heading-aware chunks and keeps a note-level summary vector alongside them. The summary is built from the first 400 words as a second, coarser representation. Four capabilities are kept separate rather than merged behind one ranking function: semantic retrieval, lexical search, graph navigation, and note operations. Scoring combines chunk and summary vectors with a max operation. Every semantic hit reports which representation determined its rank and, when the note has chunks, the most relevant passage from that note. The project includes configuration for embeddings, chunking, and summary weight, and documents benchmarks and known limitations.