This paper presents a low-cost, highly integrated dexterous robotic hand system, the core of which is the design of a structure-function integrated soft tactile fingertip module. The hand with 21 degrees of freedom (DOFs) adopts tendon driven mechanism and can simulate the complex grasping and interactive movements of the human hand. The tip of each finger is composed of a custom-designed dual-chamber barometric flexible tactile sensor and a sensor circuit board which also act as finger bone, offering the ability to detect contact forces (approximately 0.01 N). The sensor’s dynamic response characteristics, linearity, consistency, and long-term stability are evaluated through tests. As for control strategy, a pressure-feedback-based active force control grasping method is proposed that achieves closed-loop control of the contact force, suitable for manipulation of soft and deformable objects. The proposed hand has been experimentally verified in multiple task scenarios, including object slip monitoring, stiffness detection, and hard inclusion depth detection in simulated medical palpation tasks. In the palpation experiment, by combining a neural network based on the Transformer architecture to classify and identify inclusions at different depths, the system achieved an average recognition accuracy of over 95% under four embedding depth conditions, demonstrating its potential for application in medical assistance scenarios. The proposed dexterous robotic hand achieves a good balance between structural design, tactile perception, control strategy, and cost control. It has good customization and engineering application prospects and can provide reliable end-effector and tactile perception solutions for tasks such as human-robot collaboration, intelligent grasping, and robotic palpation.