Raptor code can improve the problem of error leveling by cascading precoding, but the complexity of coding is relatively high. In order to reduce the complexity of concatenated Raptor codes, this paper proposes an optimal design method based on mutual information maximization. This method is based on the gradual convergence analysis of the external information transfer graph. From the perspective of maximizing mutual information, the mathematical model of the degree distribution of check nodes under the constraint of the limited number of iterations is obtained. The simulation results show that when the number of decoding iterations is fixed, the designed Raptor code based on mutual information maximization has better BER performance than the designed Raptor code based on rate maximization.