HQE MCP-PMTs from NNVT manifest a long tail in the charge distribution. The tail component influences the estimated number of real PEs and energy resolution. In this work we tested several 8 inch MCP-PMTs to characterize the charge distributions. Gaussian mixture distribution is applied to model the charge distribution with long tail component. The parameterized charge model is used to simulate toyMC waveforms dataset. We combine poisson distribution and parameterized charge model to caculate the probability of PEs list and sample using Gibbs MCMC. The method outputs the distribution of PEs list and time of event. Maximum likelihood method is used to estimate the expect PEs of event. The result shows a significant boost in terms of resolution of PEs and time than other methods with a small bias for the charge distribution with long tail.