<p>This paper studies the challenging problem of model-free adaptive (MFA) security tracking control for nonlinear multi-agent systems (MASs) under mixed denial-of-service (DoS) attacks. First, in contrast to existing results focusing on DoS attacks with one monotonic characteristic only, a more realistic mixed DoS attacks model is constructed, which can describe multiple types of DoS attacks and reflect the real attack strategy. Second, to mitigate the negative effect of mixed DoS attacks on control performance, an effective memory event-triggered mechanism is considered. Compared with existing event-triggered schemes, the developed memory event-triggered scheme utilizes historically triggered data and allows the released data to adjust adaptively using the long-term changes of the system state, which optimizes the utilization of communication resources and withstands the effect of mixed DoS attacks. Further, with the help of a dynamic linearization technique based on memory event-triggered strategy, a linearized data model of the MASs is first established only depending on input/output information. Then, an improved memory event-triggered MFA security tracking control scheme is developed so that MASs can guarantee the tracking errors of all agents are bounded under mixed DoS attacks. Finally, a simulation example is presented of the designed MFA security tracking control method to illustrate its usefulness and advantages.</p>