localgenius 0.8.0

A personal MCP server that acts as a RAG source for your LLM

LocalGenius 🧠

A personal MCP (Model Context Protocol) server that acts as a RAG (Retrieval-Augmented Generation) source for your LLM. Seamlessly integrates with Claude Desktop to give your AI access to your local documents.

✨ Features

• 🔍 Semantic Search - Find relevant content using vector similarity
• 🤖 RAG-Powered Q&A - AI answers based on your documents using GPT models
• 📁 Local Vector Store - Fast SQLite + FAISS storage, all on your machine
• 🖥️ Claude Desktop Integration - Native MCP support with automatic setup
• 🎨 Rich CLI - Beautiful command-line interface with onboarding wizard
• ⚡ Fast Indexing - Efficient document processing and retrieval
• 🌐 Web Admin Interface - Modern React dashboard for management
• 🔒 Security First - Environment variables, no sensitive data in config files
• 📊 Multiple File Types - Support for text, markdown, code, and configuration files

🚀 Quick Start

1. Installation

# Basic installation
pip install localgenius

# With document conversion support (PDF, DOCX, PPTX, XLSX, EPUB)
pip install localgenius[documents]

# Verify installation
localgenius --help

2. Environment Setup

# Required: Set your OpenAI API key
export OPENAI_API_KEY="sk-..."

# Optional: Add to your shell profile for persistence
echo 'export OPENAI_API_KEY="sk-..."' >> ~/.bashrc  # or ~/.zshrc

3. Initialize LocalGenius

# Interactive setup wizard
localgenius init

The setup wizard will:

• Create configuration directory (~/.localgenius/)
• Set up your first data source
• Configure embedding and MCP settings
• Test your OpenAI API key

4. Add Your Documents

# Add a document directory
localgenius add-source /path/to/documents --name "My Docs"

# Add with automatic indexing
localgenius add-source /path/to/code --name "My Code" --index

5. Index Your Documents

# Index all configured sources
localgenius index

# View indexing progress and statistics
localgenius index --show

6. Test It Out

# Search your documents
localgenius search "Python examples"

# Ask AI questions about your documents
localgenius ask "What are the main topics in my documents?"

# Check system status
localgenius status

🖥️ Claude Desktop Integration

Automatic Setup (Recommended)

# Automatically configure Claude Desktop
localgenius install --claude

This command will:

• ✅ Create MCP server launcher script
• ✅ Configure Claude Desktop automatically
• ✅ Back up existing configuration
• ✅ Show you next steps

Then restart Claude Desktop completely (Cmd+Q on Mac) and look for the MCP icon (🧩)!

Manual Setup

If automatic setup doesn't work, add this to your Claude Desktop config file:

Location: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "localgenius": {
      "command": "/path/to/your/python",
      "args": ["-m", "localgenius.mcp.fastmcp_server"],
      "env": {
        "OPENAI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Using LocalGenius in Claude

Once configured, you can ask Claude to use your documents:

• "Search my documents for information about authentication"
• "What do my files say about the API architecture?"
• "Show me the LocalGenius status"
• "Based on my indexed documents, explain how to deploy this project"

🌐 Web Admin Interface

LocalGenius includes a modern React-based admin interface:

# Start the admin interface
localgenius serve --admin

Opens at: http://localhost:3000

Admin Features

• 📊 Dashboard - System status, statistics, and recent activity
• 📁 Data Sources - Add, remove, and manage document directories
• 🔍 Search & Test - Test semantic search and RAG queries live
• ⚙️ Settings - Configure embedding, chunking, and MCP parameters
• 📈 Index Management - Detailed statistics and re-indexing controls
• 🎮 Playground - Interactive testing environment for prompts and settings

📚 CLI Commands Reference

Core Commands

# Initialization and setup
localgenius init                    # First-time setup wizard
localgenius install --claude        # Configure Claude Desktop
localgenius sync-env                # Sync with environment variables

# Data source management
localgenius add-source PATH --name "Name" [--index]
localgenius remove-source PATH
localgenius list-sources

# Document indexing
localgenius index                   # Index all sources
localgenius index --source PATH     # Index specific source
localgenius index --force          # Force complete re-index
localgenius index --show           # Show detailed statistics

# Document conversion (requires localgenius[documents])
localgenius convert /path/to/docs   # Convert documents to markdown
localgenius convert --clear-cache   # Clear conversion cache
localgenius convert --show-stats    # Show cache statistics

# Search and Q&A
localgenius search "query" [--limit 5] [--threshold 0.7]
localgenius ask "question" [--model gpt-4] [--stream]

# Server operations
localgenius serve                   # MCP server (default)
localgenius serve --admin          # Web admin interface
localgenius serve --mcp            # MCP server explicitly

# System information
localgenius status                  # Show system status and statistics

Environment Variables

LocalGenius prioritizes environment variables for security:

# Required
OPENAI_API_KEY=sk-...

# Optional customization
LOCALGENIUS_DB_PATH=/custom/path/db
LOCALGENIUS_MCP_HOST=localhost
LOCALGENIUS_MCP_PORT=8765
LOCALGENIUS_MAX_CONTEXT_ITEMS=10
LOCALGENIUS_SIMILARITY_THRESHOLD=0.6

# Document conversion settings
LOCALGENIUS_CONVERT_ENABLED=true
LOCALGENIUS_CONVERT_GPU=false
LOCALGENIUS_CONVERT_MAX_SIZE_MB=100
LOCALGENIUS_CONVERT_FORMATS=pdf,docx,pptx,xlsx,epub

🔧 Available MCP Tools

LocalGenius provides these tools to Claude Desktop:

1. search - Semantic Search

Find relevant content based on similarity:

- query: Your search query
- limit: Maximum results (default: 5)  
- threshold: Similarity threshold (default: 0.7)

2. ask - RAG-Powered Q&A

Get AI-generated answers with source citations:

- question: Your question
- model: GPT model to use (default: gpt-3.5-turbo)

3. status - System Information

View index statistics and system health:

- Returns: Document count, sources, file types, and configuration

📁 Supported File Types

LocalGenius automatically detects and processes:

Text & Documentation:

• .txt, .md, .markdown, .rst
• .log, .csv, .json, .xml, .yaml, .yml

Documents (with localgenius[documents]):

• .pdf - PDF documents (using Marker or fallback)
• .docx - Word documents
• .pptx - PowerPoint presentations
• .xlsx - Excel spreadsheets
• .epub - eBooks

Code Files:

• .py, .js, .ts, .jsx, .tsx
• .java, .cpp, .c, .h, .hpp
• .cs, .php, .rb, .go, .rs
• .swift, .kt, .scala, .clj
• .sh, .bat, .ps1

Web & Markup:

• .html, .css, .scss

🛠️ Configuration

LocalGenius stores configuration in ~/.localgenius/config.yaml. Key settings:

# Embedding configuration
embedding:
  model: text-embedding-3-small
  chunk_size: 1000
  chunk_overlap: 200
  batch_size: 100

# MCP server settings  
mcp:
  host: localhost
  port: 8765
  similarity_threshold: 0.6
  max_context_items: 10

# Database settings
database:
  path: ~/.localgenius/db/localgenius.db
  vector_dimension: 1536

🔒 Security Features

• Environment Variable Priority: API keys read from environment, never stored in config files
• Local Storage: All data stays on your machine
• No Cloud Dependencies: Works entirely offline after initial setup
• Config File Safety: Sensitive data automatically excluded from config files

🚨 Troubleshooting

Installation Issues

Command not found:

pip install --upgrade localgenius
# or for development:
pip install -e .

Permission errors:

pip install --user localgenius

API Key Issues

"Invalid OpenAI API key":

# Verify your key is set correctly
echo $OPENAI_API_KEY

# Sync configuration  
localgenius sync-env

# Test with a simple command
localgenius status

Claude Desktop Integration

LocalGenius not appearing in Claude:

1. Run localgenius install --claude for automatic setup
2. Completely restart Claude Desktop (Cmd+Q, then reopen)
3. Check for the MCP icon (🧩) in Claude's interface
4. Check Console.app for any error messages

MCP connection issues:

# Test MCP server manually
localgenius serve --mcp

# Verify configuration
localgenius status

Indexing Issues

"No documents found":

# Check your data sources
localgenius list-sources

# Verify paths exist and are readable
ls -la /path/to/your/documents

# Check file types are supported
localgenius index --show

Slow indexing:

• Reduce batch size in settings
• Exclude large binary files
• Use --source to index specific directories

Performance Issues

Search too slow:

• Increase similarity threshold
• Reduce max_context_items
• Check database size with localgenius status

Memory usage:

• Reduce chunk_size and batch_size in configuration
• Consider splitting large document collections

📋 Requirements

• Python: 3.8 or higher
• OpenAI API Key: Required for embeddings and RAG functionality
• Storage: ~100MB for typical document collections
• Memory: 512MB+ recommended for large document sets

🤝 Contributing

We welcome contributions! See CONTRIBUTE.md for guidelines on:

• Setting up the development environment
• Code style and testing requirements
• Submitting issues and pull requests
• Feature request process

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Built with FastMCP for MCP server functionality
• Uses OpenAI for embeddings and language model capabilities
• Vector search powered by FAISS
• CLI interface built with Click and Rich

---

Star ⭐ this repo if LocalGenius helps you organize and search your documents with AI!