Camouflaged object detection (COD) aims to identify objects that are visually blended into their surroundings. While depth maps are often used to enhance spatial understanding and generalization for COD, existing methods struggle with poor depth quality, limited feature extraction due to reliance on convolution or Transformer backbones, and inefficient cross-modal fusion with high computational cost. To address these issues, we propose MambaCOD, a novel RGB-D COD framework based on a Cross-modal Mamba Fusion Network with Adapter Tuning. Specifically, we first introduce the Camouflaged Cognitive Visual Adapter (Cona), which works with a frozen dual-stream VMamba backbone to extract effective RGB and depth features while preserving pretrained knowledge. Second, we design a Cross State Space Model (Cross-SSM) module that integrates a well-designed Shell-Like Scan (SLS) strategy and a Dual-SSM structure for efficient cross-modal fusion. Finally, an Edge Extraction Module (EEM) and a Decoder are incorporated to enhance edge awareness and multi-scale prediction. Extensive experiments on four benchmark datasets demonstrate that MambaCOD achieves state-of-the-art performance. Our codes will be available at: https://github.com/TomorrowJW/MambaCOD .