Three-dimensional contact force perception is crucial for dexterous robotic manipulation and human-robot interaction. However, most existing tactile sensors are limited to single-axis force measurement, failing to meet complex interaction demands. To address this challenge, a magnetic-field-based flexible tactile sensor （MFFTS）was proposed for high-precision 3D contact force measurement. A grasping control strategy with force feedback was also explored. It was demonstrated through finite element simulation that the hollow pyramid structure exhibited superior overall performance among the six flexible substrate designs, with a normal sensitivity of 1183 μT/N and a tangential sensitivity of 113.10 μT/N. This structure was selected for 3D contact force analysis and the MFFTS multi-axis force decoupling estimation was investigated. Magnetic-mechanical coupling simulations revealed the coupling effects of different directional forces on MFFTS output. A backpropagation neural network （BPNN）was used to develop a decoupling method for 3D force estimation. The BPNN predictions were compared with test data, showing high measurement accuracy. The mean absolute errors for normal and tangential forces were 0.4049 N and 0.1111 N, respectively. The root mean square errors were 0.5112 N and 0.1531 N. The MFFTS was integrated onto a robotic gripper for force-controlled grasping. A multi-axis force feedback control framework was established. Experiments demonstrated that the MFFTS could monitor contact states in real time, improving grasping stability. The sensor’s simple structure facilitates fabrication, providing a potential solution for large-scale integration of 3D force sensors in robotic systems.