This paper focuses on the distributed model predictive control for multi-agent systems subject to packet loss and actuator saturation. The multi-agent systems, which can be decoupled into several subsystems, have a coupled global cost function. By decoupling the global cost function, the distributed model predictive control for multi-agent systems is approximately cast into centralized model predictive control for each agent with a cost function including the states of the neighboring agents. The exchange of states between adjacent agents realized by communication network may subject to packet loss which is assumed to obey a Bernoulli distribution with probability being known. Actuator saturation is also considered and dealt with by the Nth-step set invariance approach. To ensure the multi-agent systems is asymptotically stable, a compatibility condition is provided. The problem of controller design for agent is converted into a Linear matrix inequality optimization problem involving compatibility constraint. Furthermore, a control algorithm is obtained to ensure the asymptotic stability of the global closed-loop system as well as the recursive feasibility of the optimization problem. Finally according to a simulation example, it is concluded that the presented method is effective.