URDF-Anything is an end-to-end MLLM-based framework that jointly optimizes geometric segmentation and kinematic prediction for articulated object reconstruction.