Most of traditional robot control methods depends on the accurate model. With the complexity of robots, this kind of control methods has fallen under many restricts. On the basis of the newest research achievements in the world, a new distributed non-iterative and parallel control algorithm based on multi-agent is presented in this paper. The algorithm inherits the ideas about multi-agent and it is implemented more easily in running concurrently. It also absorbs the advantages of Incremental Direct Kinematics to make the calculations more simply. It uses the control strategy in which the distance between the manipulator end and the target could be decreased monotonously, and the corresponding non-iterative formula is deducted. Aiming at problems in manipulator control algorithms based on multi-agent, the algorithm eliminates the oscillation in the initial trajectory and essentially avoids the problem of convergent speed caused by conventional iterative algorithms.