Mid-infrared Optical Coherence Tomography (MIR-OCT) is a promising Non-Destructive Testing (NDT) technique due to its high-resolution imaging capabilities and extensive applicability across various industrial domains. The CeraMIRScan dataset comprises 29 volumes corresponding to MIR-OCT scans of 3D printed ceramic pieces and has been carefully curated to support the development of Deep Learning models for defect segmentation. Of these, 22 volumes include bounding-box annotations to enable defect localisation and classification, while all volumes are accompanied by manually generated binary segmentation masks. In total, the dataset contains 21,882 individual scans, of which 41.38% exhibit detectable defects. The dataset is organised into three primary components. The first images/ contains the 29 MIR-OCT volumes. The second annotations/raw_labels/ provides bounding-box coordinates and defect-level labels for 22 volumes. The third annotations/masks/ includes pixel-wise segmentation masks for all volumes. Image and mask files follow the naming convention 'VolumeName_SlideNumber.png', and bounding-box annotations are stored as 'VolumeName.csv'.