Abstract: Lung auscultation is crucial for early respiratory disease diagnosis. However, limited resources hinder accurate and timely assessment in many regions. In this article, we present LungScope, an intelligent embedded system designed for realtime lung sound classification. We first introduce LungLite, a lightweight classification model optimized based on our previous work, targeting deployment in resource-constrained environments. The architecture adopts redesigned LungLite blocks to reduce computational complexity while maintaining accuracy. In addition, it integrates advanced attention modules, such as SimAM and CBAM, to further enhance classification accuracy. LungLite was evaluated on the SPRSound dataset, achieving SC scores of 0.7008 for the three-class classification task and 0.5657 for the five-class classification task, with only 2.984M parameters and 0.494G FLOPs. LungLite is further integrated into LungScope by deploying it on a Raspberry Pi 4 Model B (Pi4B) with a custom-designed expansion circuit board. This integration enables optimized control, real-time lung sound acquisition, classification, and result display. The proposed portable embedded system provides an effective solution for real-time lung sound classification, supporting the basic service of healthcare in resource-limited urban settings.