TorchCP is a Python toolbox for conformal prediction research on deep learning models. It contains various implementations for posthoc and training methods for classification and regression tasks (including multi-dimension output). TorchCP is built on PyTorch (Paszke et al., 2019) and leverages the advantages of matrix computation to provide concise and efficient inference implementations. The code is licensed under the LGPL license and is open-sourced at $\href{https://github.com/ml-stat-Sustech/TorchCP}{\text{this https URL}}$.