mcp-agent-ot

# 🧠 mcp-agent-ot **Multi-Agent Context Processor** for Operational Technology (OT) data. Connects to industrial data sources (MQTT, OPC UA, Modbus), filters signals, embeds context into vector stores, and enables lightweight LLM-based Q&A — all using modular agents and Redis pub/sub. --- ## 🔧 Features - 🛰️ Plug in **MQTT**, **OPC UA**, or **Modbus** sources - 🔁 Agents run asynchronously (filtering, vectorizing, answering) - 📦 Vector DB per zone using **FAISS** - 🧠 Q&A using **MiniLM** (via `sentence-transformers`) - 🌐 FastAPI dashboard to monitor messages and triggers - 🧩 Modular, extensible, and modern Python packaging (`pyproject.toml`) --- ## 🚀 Quick Start ### 1. Clone the Repo ### 2. Install Dependencies ### 3. Start the system ### 4. Start the dashboard - Open browser: http://localhost:8000 🔌 Configuring Data Sources Edit mcpai/configs/data_sources.yml: 🧱 Architecture ingest_agent.py – connects to all OT sources dynamically filter_agent.py – identifies meaningful anomalies (e.g., "high temp") vector_agent.py – builds FAISS DB per zone llm_agent.py – answers based on vector context coordinator.py – starts everything dashboard.py – FastAPI UI redis_pubsub.py – pub/sub messaging layer 📄 License MIT 🤝 Contributing PRs and issues welcome. Designed to be extensible for edge computing and OT AI use cases.