Smart Connections MCP Server
An MCP server that gives Claude semantic search over an Obsidian vault by reusing the embeddings the Smart Connections Obsidian plugin already generated, running the same embedding model locally so a query never leaves the machine.
Smart Connections MCP Server loads the embedding vectors that the Smart Connections Obsidian plugin has already generated for every note and block in a vault, then embeds an incoming query with the same local model to rank results by cosine similarity. Its tools search across one or several vaults at both the whole-note and individual-section level, find notes similar to a given note, walk similarity links outward to map related ideas, and read a note's content or specific blocks.
If the local embedding model cannot load, for example on a very first run with no network, or if a vault has no embeddings yet, search falls back to literal keyword matching and reports that it has done so explicitly, ranking any such fallback results after true semantic matches when a search spans multiple vaults in mixed states.
It requires Node.js 20 or later, an Obsidian vault with Smart Connections embeddings already generated, and an MCP client such as Claude Desktop or Claude Code; it is written in TypeScript and licensed under MIT.