Summary
The embedding model runs on your machine, so private notes stay private.
That is the reason to pick this over syncing notes into a hosted index: nothing is uploaded, there is no key to manage, and search still works on meaning rather than exact words. You do have to run the indexing step first, and it is a manual ask rather than a background job.
What it is
A server that indexes Apple Notes into a local vector store and lets an assistant search and reference them during a conversation.
What you get
- Semantic search over your notes using the all-MiniLM-L6-v2 embeddings model, running on-device
- Full-text search alongside it
- Vector storage in LanceDB, kept locally
- Native Apple Notes access through JXA — no export step
- Storing notes back into Notes from the conversation
- Fully local execution, with no API keys involved
Requirements
macOS with Apple Notes, Bun, and a clone of the repository — point your client at bun and the path to index.ts. Ask the assistant to index your notes once before searching. Notes come back in HTML today, so formatting is rough around the edges.
Setup effort
One command — npx -y apple-notes-mcp