Abstract: Chronic stress is a major global health concern that is frequently evaluated subjectively using faulty self-reports. A Multimodal Stress Detection System for non-invasive, objective, and near-real-time assessment is presented in this study. The system uses three input modalities: Voice Tone Analysis for acoustic features using Librosa ; Text Sentiment Analysis on user interview responses using a Transformer-based model; and Facial Expression Recognition for emotional state using Face-API.js. Using a Weighted Decision-Level Fusion method, the results from various modalities are combined to provide a single stress score that ranges from 0 to 12 and is divided into six levels. Based on the identified stress level, the MSDS provides tailored advice such as exercise or meditation. The system provides a comprehensive and scalable full-stack solution for individual intervention and mental health monitoring.