NEXUS: An MCP-Based Personal Productivity Agent with Dual-LLM Failover and Human-in-the-Loop Safety

Large Language Model agents have made it possibleto talk with digital tools using everyday language but creatingsystems that are dependable and safe requires good organizationof models, integration of tools in parts and careful control ofhow the system runs. This paper introduces NEXUS, a personalproductivity agent that uses the Model Context Protocol (MCP)to carry out tasks that need steps through natural-languageconversation. NEXUS uses Google Gemini as its LLM anduses Groq-hosted Llama 3.3 70B as a backup model givinga second way to think when the main model fails. NEXUSfollows a Think–Act–Observe routine: first NEXUS reads theuser’s request then NEXUS decides what action is needed thenNEXUS asks the tool via MCP then NEXUS checks the outcomeand finally NEXUS writes the response. NEXUS works withFilesystem, To-Do and Web Search MCP services. The customTo-Do service stores task data in SQLite lets users manage taskskeeps a log of changes and allows undo actions. NEXUS also hasa Human-in-the-Loop feature that asks the user to confirm beforeany change that alters the system state is carried out. NEXUScan also start tasks on its own. Can communicate with outsidesystems through its built-in services. By keeping the thinking partof the LLM separate, from the tool actions and by using modelbackup and strict tool controls NEXUS offers a system that’smodular, dependable and able to grow for personal productivity.