Adaptive real-time applications that respond to users’ cognitive and emotional states have emerged as a critical research frontier in brain-computer interface (BCI) technology. However, existing solutions often fail to provide a cohesive toolkit that seamlessly bridges EEG signal processing, cognitive state detection, and practical software adaptation. To address this gap, we developed a structured research methodology involving EEG data collection, signal preprocessing, multi-state classification, and iterative model refinement. In controlled experiments, participants performed tasks specifically designed to induce boredom, flow, and frustration; the resulting EEG data provided the foundation for training machine learning models that could accurately recognize these states in real time. Based on these findings, we propose a flexible and modular BCI framework for Unity-based environments. This framework unites robust EEG-driven state detection with adaptive software design and was validated through two proof-of-concept applications: a game featuring dynamic difficulty adjustment and a vigilance support system for high-stakes professions.