Archive-Agent

Archive Agent is an MCP-enabled desktop-grade file indexing and search server that brings Retrieval-Augmented Generation (RAG) capabilities to your local documents. It ingests a wide variety of file types, runs OCR on images, and stores embeddings in a local vector store for fast semantic search. Through MCP, you can run natural language queries against your indexed data, ask questions, and receive structured, actionable responses directly in the CLI, GUI, or via the MCP interface. The server is designed to work with multiple AI providers (OpenAI, OpenRouter, Ollama, LM Studio) and can be configured to your privacy and performance needs. You can also tailor the indexer’s behavior, such as chunking strategy, embedding model, and OCR settings, to fit your workflow.

Prerequisites:

• Node.js (and npm) installed on your system
• Access to an AI provider API key (e.g., OpenAI, or OpenRouter) if you plan to query data

Installation steps:

1. Ensure Node.js and npm are installed:
   • macOS/Linux: curl -fsSL https://deb.nodesource.com/setup_18.x | bash -s -- && sudo apt-get install -y nodejs
   • Windows: install Node.js via the official installer from nodejs.org
2. Install Archive Agent via npx (no global install required):
   • npm install -g npm@latest
   • npx -y archive-agent
3. Alternatively, install as a local npm package (if provided by the project):
   • npm install archive-agent
4. Run the MCP server (see mcp_config for example):
   • Use the command shown in the mcp_config to start the server and expose the MCP endpoint
5. Configure your environment variables (examples):
   • OPENAI_API_KEY=your-key
   • OPENROUTER_API_KEY=your-key
   • OLLAMA_HOST=http://localhost:11434 (if using Ollama)
   • LM_STUDIO_HOST=http://localhost:9000 (if using LM Studio)

Note: Replace placeholder values with actual keys and host URLs as needed.

Tips and common issues:\n- If you switch AI providers, ensure the corresponding API keys and host URLs are correctly set in environment variables.\n- For best privacy, consider local LLM options (Ollama or LM Studio) and ensure your hardware can support the model sizes you select.\n- The local Qdrant vector store will hold your embeddings; ensure you have adequate disk space for your corpus.\n- If you encounter connectivity issues to the MCP interface, verify that your MCP endpoint is reachable from your client and that any firewalls allow the required ports.\n- The OCR feature is experimental; for best results, provide high-quality images and consider pre-processing where appropriate.\n- Review the Archive Agent settings and profile configurations to tailor file tracking patterns and chunking behavior to your documents.