An innovative multi-level shape memory alloy (SMA) cable damper is proposed for bridges in near-fault seismic regions. This study commences with the introduction of a constitutive material model for simulating SMA cables. The proposed model not only considers the self-centering hysteretic property but also accounts for the strength degradation and residual strain accumulation effects. Subsequently, the layout and working mechanism of the proposed multi-level SMA cable damper is presented. A seismic fragility analysis framework involving Box-Cox transformation and Bayesian inference is subsequently presented. The Box-Cox transformation is utilized to establish the nonlinear probabilistic seismic demand models (PSDMs), and the Bayesian-based logistic regression is employed to formulate fragility functions. Finally, a three-span continuous reinforced concrete (RC) girder bridge incorporating the multi-level SMA dampers is selected as an example to demonstrate the proposed framework. A total of 160 near-fault ground motions, containing pulse and non-pulse motions, are applied along the longitudinal direction of the bridge. To investigate the influences of different parameter values of the SMA cable damper on fragility functions, various parameters are generated with respect to a central composite design philosophy. Based on the seismic fragility analysis, it is found that the SMA isolation damper reduces the damage probabilities of the bearing and pier for all four damage states. In addition, the failure probability of the bearing decreases with the increase of the axial stiffness of the SMA damper, which is opposite to the pier. Moreover, the optimum designed SMA damper can significantly reduce the damage probability of the case study bridge, and the parameter values of the damper should be carefully selected to achieve the desired seismic performance.