The manipulator grasping system based on the local visual system is only be applicable to the static and already existing in the visual training model parts and can’t accurately identify and grasp new or motional parts. In this paper, a part grasping system for manipulator in cloud environment is designed. By using the excellent storage performance and efficient computing capacity of cloud, the types of parts that can be grabbed by manipulator are enriched, the calculation speed of part data is improved, and the dynamic grabbing is realized. Under the structure of cloud-edge-terminal, using the part type, running speed, height, spatial position and angle information stored in the cloud, combined with the robot running speed and Rodriguez formula, the manipulator TCP position and pose of the part grabbing point is calculated. The capture strategy of parts is generated in the cloud. By the network connection between the upper computer and the manipulator, the control command generated online is transmitted to the manipulator to complete the dynamic grasping spare parts. At the same time, the status data of the manipulator is obtained and analyzed during the grasping process. The method proposed in this paper improves the adaptability of the part grabbing system, and provides a technical basis for motional part grabbing in the cloud environment.