Integrating Model Context Protocol and Agentic AI for Robust Cyberbullying Detection on Social Media

Abstract

This study presents a robust framework for addressing security challenges in Agentic AI systems through contextual protocol-driven deep learning architectures. The proposed hybrid model integrates NLP-based text preprocessing, CNN-driven image feature extraction, and transformer-based fusion layers, enabling dynamic contextual adaptation. Using real-world social media data, the model achieved an average accuracy of 90%, precision of 91%, and recall of 89%, outperforming conventional CNN and RNN frameworks by over 8% in overall efficiency. The hybridization of transformer fusion and fully connected deep layers reduced inference latency by 13%, ensuring scalable real-time defense against data manipulation and phishing-based adversarial attacks. Performance evaluation and comparative analysis (Figs. 3–5) confirm that contextual learning protocols substantially enhance model resilience, explainability, and energy efficiency within secure AI ecosystems.