NeuSpell is a neural spelling correction toolkit that leverages pre-trained transformer models including BERT, DistilBERT, and XLM-RoBERTa for spell correction tasks. The toolkit provides APIs for fine-tuning models on custom data, generating synthetic training data through text noising, and evaluating spelling correction performance. It was accepted as a system demonstration at EMNLP 2020 and offers pre-trained checkpoints available via Hugging Face.