Attributed graphs are central to modeling relational data in domains such as fraud analytics, healthcare, and social networks. Detecting anomalies in these settings is difficult because irregularities may reside in higher-order interaction patterns (motifs) and emerge differently across multiple attribute views. Existing methods typically (i) under-model motif-centric irregularities, obscuring anomalies embedded in higher-order structures, and (ii) insufficiently integrate complementary information across views, weakening anomaly identification. To address these challenges, we propose MAD-MG, a novel Motif-Augmented Anomaly Detection framework for Multi-View Attributed Graphs. First, MAD-MG transforms multi-view attributed graphs into motif-enriched representations by partitioning them into view-specific subgraphs, systematically extracting motifs, and augmenting them with virtual motif nodes, thereby embedding higher-order structural semantics while preserving the distinct attribute context of each perspective. Second, a self-attention-driven cross-view integration mechanism emphasizes critical inter-view and intra-view correlations, yielding refined embeddings that are further consolidated through motif feature aggregation. Finally, anomalies are identified by jointly reconstructing structure and attributes, allowing subtle yet structurally significant irregularities to be captured. Extensive experiments on benchmark datasets demonstrate that MAD-MG consistently outperforms state-of-the-art methods in precision, recall, F1-score, and AUC, establishing a new benchmark for scalable and adaptive motif-level anomaly detection in multi-view attributed graphs.