📊 Usage Examples

Document Seeding

# Seed documents for specific client
npm run seed -- --client work --filesdir /path/to/work/documents

# Force overwrite existing data (full reprocessing)
npm run seed -- --client personal --filesdir /path/to/personal/docs --overwrite

# Validate documents without seeding  
npm run seed -- --client research --filesdir /path/to/research/docs --validate-only

# Debug mode for troubleshooting
npm run seed -- --client client_a --filesdir /path/to/docs --debug

MCP Server Usage

# Run the MCP server
npm start

# Or in development mode with watch
npm run watch

Claude Desktop Integration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "qdrant-rag": {
      "command": "node",
      "args": ["/absolute/path/to/qdrant-mcp-hybrid/dist/index.js"],
      "env": {
        "QDRANT_URL": "http://localhost:6333",
        "QDRANT_API_KEY": "your-api-key-if-needed",
        "CLIENT_COLLECTIONS": "work,personal,research"
      }
    }
  }
}

🔧 Available MCP Tools

`collection_info`

Get status of all collections and clients.

// No parameters needed
collection_info()
// Returns: Collection stats, client list, system status

`catalog_search`

Search document summaries for a specific client.

{
  "query": "quarterly business strategy",
  "client": "work", 
  "limit": 10
}

`chunks_search`

Search document chunks with optional source filtering.

{
  "query": "machine learning implementation",
  "client": "research",
  "source": "/path/to/specific/document.md",  // optional
  "limit": 5
}

`all_chunks_search`

Search across all clients and collections.

{
  "query": "project management best practices",
  "limit": 20
}

🏗️ Architecture Deep Dive

Collection Structure

Qdrant Collections:
├── work_catalog           # Document summaries for work
├── work_chunks            # Document chunks for work  
├── personal_catalog       # Document summaries for personal
├── personal_chunks        # Document chunks for personal
├── research_catalog       # Document summaries for research
├── research_chunks        # Document chunks for research
└── ... (per client)

Data Flow Pipeline

Documents → Hash Check → Content Extract → LM Summary → 
Chunk Split → BGE-M3 Embed → Batch Process → Qdrant Store → MCP Search

Document Processing Pipeline

Directory Scan - Find all supported documents (.pdf, .md, .txt, .docx)
Hash Validation - SHA256 deduplication (skip unchanged files)
Content Processing - Extract text using appropriate parsers
Summary Generation - LM Studio Qwen3 creates document overviews
Chunk Creation - Split documents with configurable overlap
Batch Embedding - BGE-M3 vectorization in efficient batches

Categories

Current Category: claude-compliant

qdrant-mcp-hybrid-public