Published January 1, 2026 | Version v1

Nunnarri: An AI-Enabled Universal IoT Management Platform Integrating MQTT, Local Language Models, and Model Context Protocol

Description

The growing use of heterogeneous Internet of Things (IoT) devices creates operational challenges in device onboarding, telemetry collection, control, data persistence, and integration with intelligent services. This paper presents Nunnarri, a modular IoT management platform designed to bring these functions into a single deployable system. The public implementation combines a React and Vite dashboard, Node.js and Express API services, an MQTT/WebSocket gateway, two PostgreSQL database services, an Ollama-connected AI backend, and a Model Context Protocol (MCP) server. The architecture separates identity and access data from IoT telemetry, provides a virtual-pin abstraction for device interaction, and exposes selected device-management operations to compatible AI clients through MCP tools, resources, and prompts. The system is packaged as a multi-container Docker Compose application and includes firmware integration material for ESP8266-class devices. The contribution of this work is an implementation-oriented architecture that connects conventional IoT messaging and web management with locally hosted language-model assistance and a standardized tool interface. The repository documents the intended system components; however, quantitative claims about latency, throughput, anomaly-detection quality, reliability, and resource use require controlled evaluation. A repeatable evaluation plan is therefore defined for message delivery, end-to-end command delay, service recovery, AI-assisted task completion, and security controls

Files

IJSET_V14_issue5_172.pdf

Files (243.2 kB)

Name Size
md5:0dc438d3a8d8543a2232d070362a532d
243.2 kB Preview Download

Additional details