With the rapid increase in Internet of Things, large-scale streaming data is being generated at an unprecedented rate. Identifying and handling anomalies in these data streams is crucial for improving data quality and enhancing data analysis performance. Clustering is an effective method for processing streaming data, as it groups data points to uncover patterns. This article introduces Density clustering-guided Streaming data Anomaly Detection (DeSAD) algorithm, which achieves efficient anomaly detection by dynamically learning density-based clusters as reference sets while eliminating the need for manual parameter tuning. DeSAD first uses density-based clustering techniques to identify potential noise points, and then filters out the true anomalies from these points. This process not only increases computational efficiency but also adapts to the diversity and dynamic nature of streaming data. In addition, DeSAD adjusts its clustering parameters in real-time based on the data distribution, making it highly suitable for varying types of data streams. Experimental results demonstrate its effectiveness in detecting anomalies across 11 benchmark datasets and 4 synthetic datasets, showcasing its potential in handling large-scale, dynamic data streams.