Multiple Sclerosis (MS) is an incurable chronic autoimmune disease, heavily relies on comprehensive analysis of brain lesions from MRI scans for an accurate diagnosis and progression evaluation, a task that can be time-consuming and prone to inter-rater variability among neurologists. While existing computer-aided diagnosis (CAD) systems have made significant advances in lesion detection and segmentation, they still fail to bridge the gap between imaging findings and actionable insights. To address these limitations, we propose a novel agentic AI framework that unifies multi-scale feature extraction with interpretable reasoning. With a hierarchical attention network that integrates lesion-level, region-level, and global-level features to predict pathological involvement across 48 brain regions. Building upon this multi-scale foundation, a GPT-4o-based LLM agent orchestrates region classification and symptom mapping, dynamically querying a clinical knowledge database to generate patient-specific reports enabling advanced reasoning about lesion patterns and their clinical implications. Through comprehensive evaluation on 100 patients from public datasets, our framework outperforms SOTA architectures and achieves 0.85 ± 0.04 AUC in regional classification while delivering clinician-friendly interpretations. By converting imaging data into structured clinical insights, we advance AI based MS care from passive lesion quantification into active decision support; representing a paradigm shift toward agentic CAD systems.