OBJECTIVE: Type 2 diabetes mellitus (T2DM) is a significant risk factor for mild cognitive impairment (MCI). Here, we identified a T2DM-specific effective connectivity (EC) network, the dynamic features of which could be used to distinguish T2DM patients with MCI from healthy controls (HC) and the correlation with cognitive performance. METHODS: Local and multicentered T2DM patients and matched HC who underwent functional magnetic resonance imaging were recruited. Their static and dynamic effective connectivity were compared. The relationships between connectome characteristics and cognitive performance were also evaluated. RESULTS: The nodes of the T2DM-related static causality network included the anterior central gyrus, tail of the parahippocampal gyrus, posterior superior temporal sulcus, posterior central parietal lobe, posterior central gyrus and V5 region of the occipital lobe. The V5 region of the visual cortex was the core node. In the multicentered dataset, compared with the HC group, the T2DM with MCI group had significantly greater fractional window and mean dwell time. Fractional windows of the state, which was dominated by the interaction of the nodes from SomMot_Network, Limbic_Network, Default_Network, in the T2DM-specific network increased with poorer cognitive performance in T2DM with MCI patients. CONCLUSION: Our findings provide insights into the neurobiological mechanisms of the cognitive impairment of T2DM patients from a dynamic network perspective, which may ultimately inform more targeted and effective strategies to prevent MCI.