Graph RAG MCP Server for Obsidian

A powerful local-first Graph-RAG system that combines ChromaDB unified store with metadata-based graph relationships and Gemini 2.5 Flash for intelligent Q&A over your Obsidian vault.

🌟 Features

• 📊 Unified Database Architecture: ChromaDB with metadata-based graph relationships for both semantic search and graph traversal
• 🧠 Intelligent Semantic Chunking: Respects markdown structure (headers, sections, lists, code blocks)
• 🎯 PARA Taxonomy Classification: AI-powered organization using Projects, Areas, Resources, Archive system
• 🤖 RAG-Powered Q&A: Multi-hop retrieval with Gemini 2.5 Flash
• 🕸️ Graph Navigation: Explore backlinks, tags, and semantic relationships
• 🔄 Real-time Sync: File watcher for automatic indexing
• 🏠 Local-First: All processing happens locally (except Gemini API calls)
• 📝 Multi-Client MCP Support: Works with Claude Desktop, Cursor, and Raycast
• ⚡ Dual Transport Modes: stdio for Claude Desktop, HTTP for other clients
• 🛠️ Automated Installation: One-command setup with client detection
• 🔒 Strongly Typed: Pydantic models throughout for reliability

🎯 Supported MCP Clients

• 🤖 Claude Desktop: Full stdio integration with automatic configuration
• 📝 Cursor: HTTP mode with MCP extension support
• ⚡ Raycast: HTTP API with custom extension templates
• 🔌 Any MCP Client: Standard MCP protocol support (stdio/HTTP)

🏗️ Architecture

┌─────────────────┐    ┌─────────────────┐
│   Obsidian      │    │   ChromaDB      │
│     Vault       │───▶│ (Unified Store) │
│                 │    │ Vector + Graph  │
└─────────────────┘    │   Metadata      │
         │              └─────────────────┘
         │                       │
         │                       ▼
         │              ┌─────────────────┐
         │              │      DSPy       │
         └─────────────▶│   RAG Engine    │
                        │  + Gemini 2.5   │
                        │ (Multi-hop      │
                        │  Retrieval)     │
                        └─────────────────┘
                                 │
                                 ▼
                        ┌─────────────────┐
                        │   MCP Server    │
                        │   (FastMCP)     │
                        └─────────────────┘
                                 │
                                 ▼
                        ┌─────────────────┐
                        │ Claude Desktop  │
                        │   Integration   │
                        └─────────────────┘

🚀 Quick Start

Automated Installation (Recommended)

The easiest way to get started with Claude Desktop, Cursor, or Raycast:

# Interactive setup wizard
uv run install.py

# Or non-interactive with your settings
uv run install.py --vault "/path/to/your/vault" --api-key "your_key"

The installer will:

• ✅ Detect your installed MCP clients (Claude Desktop, Cursor, Raycast)
• ⚙️ Configure each client automatically
• 📦 Install all dependencies
• 🧪 Test the installation
• 📝 Create environment configuration

Manual Installation

If you prefer manual setup:

1. Install uv (if not installed)

curl -LsSf https://astral.sh/uv/install.sh | sh  # macOS/Linux

2. Install Dependencies

uv sync

3. Configure Environment

cp configs/.env.example .env
# Edit .env and add your GEMINI_API_KEY and vault paths

4. Index Your Vault

# Full indexing (unified ChromaDB store)
uv run scripts/reindex.py all

# Check indexing status
uv run scripts/reindex.py status

5. Configure Your MCP Client

Claude Desktop (stdio mode):

{
  "mcpServers": {
    "graph-rag-obsidian": {
      "command": "uvx",
      "args": ["--python", "3.13", "--from", ".", "graph-rag-mcp-stdio"],
      "cwd": "/path/to/graph-rag-mcp-server",
      "env": {
        "GEMINI_API_KEY": "your

... [View full README on GitHub](https://github.com/ferparra/graph-rag-mcp-server#readme)