GitHub - 2dogsandanerd/DAUT: DAUT – Documentation Auto Updater - AI-powered documentation generator for your codebase. MCP-Connector

5 min read Original article ↗

📚 DAUT - Documentation Auto-Update Tool

AI-powered documentation generator that keeps your docs in sync with your code

Python 3.9+ License: MIT Streamlit

DAUT scans your codebase, detects undocumented code, and automatically generates comprehensive documentation using LLM (Ollama). Perfect for maintaining up-to-date API docs, class references, and function documentation across Python, JavaScript, and TypeScript projects.

✨ Features

  • 🔍 Universal Code Scanner - Detects functions, classes, API endpoints across Python, JS, TS
  • 🤖 AI Documentation Generation - Uses Ollama to generate human-readable docs
  • 📊 Live Progress Tracking - Real-time progress bars and statistics
  • 🎯 Smart File Detection - Respects .gitignore, skips venv/node_modules automatically
  • 💾 ChromaDB Integration - Semantic search and context-aware documentation
  • ⚡ Resume Support - Skip already-generated docs, continue where you left off
  • 🔌 MCP Server - Expose RAG capabilities to external agents (Claude, Cursor, etc.)
  • 🎨 Beautiful UI - Streamlit-based interface + powerful CLI

🧠 RAG Strategy (Under the Hood)

DAUT uses a sophisticated structural indexing approach to ensure high-quality answers:

  1. Unified Knowledge Base 🌐 All files, regardless of their folder depth, are indexed into a single project-wide collection (e.g., rag_enterprise_core_code). This prevents context fragmentation and ensures the AI sees the "Big Picture".

  2. Full-Content Embedding 📖 Unlike simple splitters that chop text into arbitrary chunks, DAUT indexes the full content of your documentation files. This preserves the complete context of tutorials and guides.

  3. Structure-Aware Code Indexing 🏗️ Code is not just text. We parse the AST (Abstract Syntax Tree) to treat Classes, Functions, and API Endpoints as distinct semantic entities.

🚀 Quick Start

Installation

# Clone and setup
git clone <your-repo>
cd doc_updater_app
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Generate Docs in 3 Steps

# Launch UI
streamlit run src/ui/main.py
  1. Select Project → Browse to your codebase
  2. Scan → Analyze code and find undocumented elements
  3. Generate → AI creates comprehensive docs

CLI Mode:

python -m src.docs_updater /path/to/project

📸 Screenshots

Scan Progress

Scanning with real-time progress

Analysis Dashboard

Diskrepancy analysis and statistics

AI Documentation Generation

Live documentation generation with Ollama

Generated Documentation Files

Auto-generated markdown documentation

Documentation Files Overview

Explorer view of generated docs

Explorer view of generated docs

🎯 Use Cases

  • API Documentation - Auto-generate REST API endpoint docs
  • Code Onboarding - Help new developers understand your codebase
  • Documentation Audits - Find and fix documentation gaps
  • Legacy Code - Document undocumented legacy systems
  • Continuous Docs - Keep docs in sync with code changes

💡 Best Practices

Ensure High-Quality RAG Results

To avoid "diluting" the AI's knowledge base with outdated information:

  • Prioritize auto_docs: These files are generated directly from the current codebase and represent the "source of truth".
  • Exclude Legacy Docs: If you have an old docs/ folder with manual (potentially outdated) documentation, consider adding docs/ to the Exclude Patterns in the Filter Management sidebar.
  • Why? If the RAG system indexes both current code (via auto_docs) and outdated manuals (via docs/), it might retrieve conflicting information. By filtering out legacy docs, you ensure a "Pure Code-Truth" knowledge base.

📋 Example Output

Input: Python function

def get_session(session_id: str):
    """Retrieve session history."""
    return db.query(session_id)

Generated Documentation:

## get_session

### Description
The `get_session` API endpoint retrieves the conversation history for
a specific session. Requires permission to view session history.

### Parameters
| Name | Type | Default |
|------|------|---------|
| session_id | str | None |

### Return Value
Returns the session history including session ID and message list.

### Example
```bash
GET /sessions/12345

Error Handling

Returns 500 on errors, 403 if permission denied.

doc_updater_app/ ├── src/ │ ├── core/ # Config management, project analysis │ ├── scanner/ # Code & documentation scanners │ ├── matcher/ # Discrepancy detection │ ├── llm/ # Ollama integration │ ├── chroma/ # ChromaDB vector store │ ├── updater/ # Documentation update engine │ └── ui/ # Streamlit interface ├── requirements.txt └── setup.py


## 🔧 Configuration

**service_config.json:**
```json
{
  "ollama_host": "http://localhost:11434",
  "chroma_host": "localhost",
  "chroma_port": 8000,
  "ollama_timeout": 120
}

🛠️ Requirements

  • Python 3.9+
  • Ollama (optional, for AI generation) 
    # Install: https://ollama.ai
ollama pull llama3
  • ChromaDB (optional, for semantic search) 
    pip install chromadb
chroma run --path ./chromadb_data --port 8000

📚 Supported Languages & Formats

Code:

  • Python (.py)
  • JavaScript/TypeScript (.js, .ts, .tsx, .jsx)

Documentation:

  • Markdown (.md)
  • reStructuredText (.rst)
  • Plain text (.txt)

🎨 Features in Detail

Smart Progress Tracking

🔍 Scanning: [45/1234] 3.6% - api_service.py

[1/150] Verarbeite: get_session (api_endpoint)
    ✅ Gespeichert: get_session.api.md
[2/150] Verarbeite: delete_session (api_endpoint)
    ⏭️  Übersprungen (existiert): delete_session.api.md

Resume Support

Stop and restart anytime - already generated docs are automatically skipped!

Diskrepanz Analysis

  • Undocumented Code - Functions/classes without docs

  • Outdated Documentation - Docs that don't match current code

  • Mismatched Elements - Signature changes, parameter updates

  • Mismatched Elements - Signature changes, parameter updates

🔌 MCP Server Integration

DAUT includes a Model Context Protocol (MCP) server, allowing you to connect external AI agents (like Claude Desktop, Cursor, or other LLMs) directly to your project's knowledge base.

Features

  • Secure Access: Protected via API Key (Bearer Token).
  • RAG Tools:
    • query_rag(query): Semantic search in your code and documentation.
    • read_documentation_file(path): Read full content of generated docs.
    • list_documentation_files(): List available documentation.
  • Monitoring: Live connection tracking via the Web UI.

🚀 Usage

Manual Start:

# Start the server (Default port: 8001)
./start_mcp.sh

Auto-Start (Systemd): Run as a background service that survives reboots:

chmod +x install_service.sh
sudo ./install_service.sh

🔐 Security & Configuration

The server requires an API Key for all requests. You MUST configure this to secure your data.

Setting the API Key:

  1. Edit start_mcp.sh (for manual start) or daut-mcp.service (for systemd).
  2. Change the MCP_API_KEY variable: 
    export MCP_API_KEY="your-secure-password-here"
  3. Restart the server.

Environment Variables:

Variable Default Description
MCP_PORT 8001 Port for the MCP SSE endpoint
MCP_HOST 0.0.0.0 Bind address
MCP_API_KEY secret-token-123 REQUIRED: Auth token for clients

Connect a Client:

  • URL: http://<your-server-ip>:8001/mcp/sse
  • Auth: Header Authorization: Bearer <your-key>

🤝 Contributing

Contributions welcome! This project is under active development.

📄 License

MIT License - see LICENSE file for details

🙏 Acknowledgments

🚦 Project Status

Current Version: 1.0.0 (Stable)

All core features implemented:

  • ✅ Universal code scanning
  • ✅ AI documentation generation
  • ✅ Progress tracking and resume support
  • ✅ ChromaDB integration
  • ✅ Streamlit UI + CLI

📞 Support

Found a bug or have a feature request? Open an issue!

Made with ❤️ for developers who love good documentation