Regular Baduanjin exercise intervention was proven to be beneficial in improving the cognitive ability and physical performance of older adults with different health conditions but was unclear to influence the structural plasticity of the hippocampus. This study aimed to explore the modulation of hippocampal subregions as a mechanism by which Baduanjin exercise improves cognitive frailty in older adults. A total of 102 community-dwelling older adults with cognitive frailty were recruited and randomly allocated to the Baduanjin exercise training group and usual physical activity control group. The participants in the Baduanjin exercise training group participated in a 24-week Baduanjin exercise intervention program with an exercise frequency of 60 min per day, 3 days per week. Cognitive ability and physical frailty were assessed, and MRI scans were performed on all participants at baseline and after 24 weeks of intervention. The structural MRI data were processed with MRIConvert (version 2.0 Rev. 235) and FreeSurfer (version 6.0.0) software. Data analyses were performed using the independent sample t tests/Mann-Whitney U tests with the Bonferroni correction, mixed linear model, correlation, or mediation analysis by the SPSS 24.0 software (IBM Corp, Armonk, NY, United States). After 24 weeks of intervention, a statistically significant increase was found for the Montreal Cognitive Assessment (MoCA) scores (p = 0.002) with a large effect size (Cohen's d = 0.94) and the significant interaction effect (P goup × time < 0.05), Memory Quotient (MQ) scores (p = 0.019) with a medium effect size (Cohen's d = 0.688) and the significant interaction effect (P goup × time < 0.05), and other parameters of WMS-RC test including pictures (p = 0.042), recognition (p = 0.017), and association (p = 0.045) test with a medium effect size (Cohens' d = 0.592, 0.703, and 0.581) for the Baduanjin training group, while significant decrease for the Edmonton Frailty Scale (EFS) score (p = 0.022), with a medium effect size (Cohen's d = -0.659) and the significant interaction effect (P goup × time < 0.05) for the Baduanjin training group. The differences in the left parasubiculum, Hippocampal Amygdala Transition Area (HATA), right Cornu Ammonis Subfield 1 (CA1) and presubiculum volumes from baseline to 24 weeks after intervention in the Baduanjin training group were significantly greater than those in the control group (p < 0.05/12). Further analysis showed that the changes in right CA1 volume were positively correlated with the changes in MoCA and MQ scores (r = 0.510, p = 0.015; r = 0.484, p = 0.022;), the changes in right presubiculum and left parasubiculum volumes were positively correlated with the changes in MQ (r = 0.435, p = 0.043) and picture test scores (r = 0.509, p = 0.016), respectively, and the changes in left parasubiculum and HATA volumes were negatively correlated with the changes in EFS scores (r = -0.534, p = 0.011; r = -0.575, p = 0.005) in the Baduanjin training group, even after adjusting for age, sex, years of education and marital status; furthermore, the volume changes in left parasubiculum and left HATA significantly mediated the Baduanjin exercise training-induced decrease in the EFS scores (β = 0.376, 95% CI 0.024 ~ 0.947; β = 0.484, 95% CI 0.091 ~ 0.995); the changes of left parasubiculum and right CA1 significantly mediated the Baduanjin exercise training-induced increase in the picture and MO scores (β = -0.83, 95% CI-1.95 ~ -0.002; β = -2.44, 95% CI-5.99 ~ -0.32). A 24-week Baduanjin exercise intervention effectively improved cognitive ability and reduced physical frailty in community-dwelling older adults with cognitive frailty, and the mechanism might be associated with modulating the structural plasticity of the hippocampal subregion.