Anomaly detection on attributed networks, which aims to identify abnormal nodes that differ from the majority within a network, has received significant attention. However, existing methods generally overlook the rich topological information that is crucial for understanding the complex patterns. To overcome these constraints, this paper proposes an innovative framework, ANTI, to enhance anomaly detection by integrating topological identity features into contrastive learning. Equipped with graphlet isomorphic counting, ANTI can explicitly capture local structural characteristics. Firstly, ANTI performs topological relation-aware community representation learning that can fully consider the topological identity of the nodes within the community, and then leverage the node-community contrastive learning. In addition, to better take advantage of the attributed and topological information, ANTI reconstructs the features from the community by leveraging decoders. Finally, the anomaly score inferences from the contrastive learning module and the reconstruction module are aggregated to form the final anomaly score. Experiments are conducted on various real-world datasets demonstrate ANTI’s enhanced performance, highlighting its capability to effectively detect anomalies through the integration of attribute and topological data. The code and demo data are available at https://github.com/haoyitong/ANTI-main.git .