Camouflage object detection (COD) aims to identify objects concealed through self-camouflage or environmental occlusion. While existing methods have made notable progress, they still face three key challenges: ineffective modeling of multi-scale long-range contextual dependencies, leading to over-segmentation or local omission; insufficient utilization of high-frequency spatial information, causing performance degradation in scenes with blurred boundaries; and large model sizes, limiting practical deployment. To address these issues, we propose MHNet, a simple yet effective multi-scale long-range attention perception and high-frequency boundary-guided network for COD. Specifically, we design: A multi-scale spatial coordinate attention module (MSCA) to extract fine-grained multi-scale spatial features and long-range contextual structures. A semantic information supplement module (SISM) to mitigate semantic context dilution and enhance feature flow. A high-frequency boundary-guided feature enhancement module (HBGM) to refine boundary details for more precise object localization. Additionally, we introduce multi-scale skip connections between decoders to aggregate object features across different scales, strengthening contextual representation and improving feature discrimination. Extensive experiments on three public benchmark datasets show that MHNet surpasses 11 state-of-the-art (SOTA) methods in both accuracy and model efficiency. The code will be available at: https://github.com/Ricardo486/q486 .