DAUT

DAUT provides an MCP (Model Context Protocol) server that exposes its RAG (retrieval-augmented generation) capabilities to external agents and LLMs. The MCP server lets clients perform semantic queries over the code and generated documentation, read full documentation files, and list available docs, all secured with an API key. To connect, run the MCP server and supply the API key with subsequent requests. Once running, external tools can query the knowledge base, read docs, and navigate the repository’s documentation surface via the exposed endpoints. The integration is designed to allow agents like Claude Desktop, Cursor, or other LLMs to leverage the DAUT knowledge base in real-time, supporting use cases such as API endpoint documentation, code onboarding, and documentation audits.

Tool capabilities typically include:

• query_rag(query): perform semantic search across code and documentation to surface relevant results.
• read_documentation_file(path): fetch the full content of a generated documentation file.
• list_documentation_files(): enumerate available generated docs for quick reference.
• Live monitoring of MCP connections via the included UI or logs, enabling visibility into active clients and queries.

To use these tools, obtain or generate an API key, start the MCP server, and configure your client to pass the key in the Authorization header (Bearer token) for each request. The server port can be customized via MCP_PORT and the key via MCP_API_KEY. Ensure Ollama and optional ChromaDB are set up if you plan to leverage DAUT’s AI generation and semantic search features on the upstream code and docs.

Prerequisites:

• Python 3.9+
• A working codebase for scanning and documenting (Python, JavaScript, and TypeScript supported)
• Optional: Ollama for AI-driven documentation generation and ChromaDB for semantic search

Step-by-step install:

1. Clone the repository and navigate to the project root

   • git clone <your-repo>
   • cd daut

2. Set up a Python virtual environment

   • python -m venv venv
   • source venv/bin/activate # macOS/Linux
   • venv\Scripts\activate # Windows

3. Install dependencies

   • pip install -r requirements.txt
   • Optionally install Ollama and start Ollama if you plan to generate docs via LLM:
     • ollama pull llama3

4. Configure environment variables (example)

   • export MCP_API_KEY="your-secure-password-here" # on Unix-like systems
   • export MCP_PORT=8001
   • (Windows) set MCP_API_KEY=your-secure-password-here
   • (Windows) set MCP_PORT=8001

5. Run the MCP server (choose one of the supported run methods)

   • Python module method (recommended if the project exposes a module):
     • python -m daut.mcp_server
   • Alternatively, use the provided startup script if available:
     • ./start_mcp.sh

6. Verify the server is running

   • Check logs for a message indicating the MCP server is listening on port 8001 (or your configured port)
   • Test with a simple curl request using the Bearer token in the Authorization header

Notes:

• You may need to install additional dependencies for optional features (e.g., ChromaDB, Ollama).
• If you customize the MCP port, ensure your clients are pointed to the new port.

Tips and common issues:

• API Key security: Always keep MCP_API_KEY in a secure location and do not commit it to version control. Use environment-specific overrides for development, staging, and production.
• Port configuration: If 8001 is in use, pick an available port and update MCP_PORT accordingly.
• Ollama integration: If you rely on AI-driven docs, ensure Ollama is installed and running (ollama start) and that you have pulled the appropriate model(s).
• Data persistence: DAUT’s RAG features assume a vector store (ChromaDB). If you disable or do not configure ChromaDB, semantic search features will be limited to exact text matching.
• Access control: The MCP server requires a Bearer token; ensure clients supply Authorization: Bearer <token>.
• Debugging: If requests fail, check server logs for missing environment variables, schema changes, or authentication errors. Ensure the module path in the mcp_config matches the actual Python module that exposes the MCP server entry point.
• Networking: When running behind a reverse proxy or in containers, ensure proper header forwarding and TLS termination if exposing the API publicly.