RAG Obsidian Server

Semantic search over your Obsidian vault via an MCP server. Uses ChromaDB for vector storage and all-MiniLM-L6-v2 for embeddings.

Prerequisites

• Python 3.13+
• uv (recommended) or pip

Setup

uv sync

Edit config.yaml to point to your vault:

vault_path: "/path/to/your/Obsidian/vault"
chroma_db_path: "./data/chroma_db"
embedding_model: "all-MiniLM-L6-v2"
collection_name: "obsidian_vault"
chunk_size: 1000
chunk_overlap: 100
top_k_default: 5

Indexing

Initial index (first run)

uv run python index.py

This walks all .md files in the vault, embeds them in chunks, and stores them in ChromaDB.

Incremental index (after adding or editing notes)

Run the same command — it skips files whose modification time hasn't changed:

uv run python index.py

Only new or modified notes are re-indexed. Deleted notes are removed from the database automatically.

Force full re-index

To re-embed everything from scratch (e.g. after changing chunk_size or embedding_model):

uv run python index.py --force

Override vault path

uv run python index.py --vault "/path/to/other/vault"

Starting the MCP Server

Run the server in stdio mode (the transport Claude Code and other MCP clients expect):

uv run python server.py

Registering with Claude Code

Add the server to your Claude Code MCP config (.claude/settings.json or global settings):

{
  "mcpServers": {
    "obsidian-vault": {
      "command": "uv",
      "args": ["run", "python", "server.py"],
      "cwd": "/path/to/RAG_Obsidian"
    }
  }
}

Available MCP Tools

Tool	Description
search_notes	Semantic search across indexed notes. Returns top-k chunks with relevance scores and obsidian:// deep links.
read_note	Read the full content of a note by its relative path (e.g. Projects/MyProject.md).
list_notes	List all notes in the vault or a subfolder.