<p>This work describes MaithiliMT, a multi-domain parallel corpus for the Hindi-Maithili language pair. Maithili is one of the 22 scheduled languages of India and is spoken by 14 million people. Maithili and Hindi belong to the Indo-Aryan language family and share lexical and syntactic similarities. We manually analyze the available Hindi-Maithili corpus and observe that the available corpus is of poor quality and that Hindi-Maithili machine translation (MT) can benefit from additional corpora. We create additional parallel corpora for administration, judicial, education, and general domains. The newly created corpus consists of a synthetic corpus created via back-translation and manual translation of monolingual Hindi data into Maithili. We also create post-edited data by manually correcting the generated synthetic data. We follow multilingual NMT, code-mixed augmentation, and automatic post-editing approaches to train baseline neural machine translation (NMT) models. The newly created dataset is of higher quality than the existing Hindi-Maithili corpus, and the NMT results show that multilingual models give better results than the other approaches, with an average of 16.4 and 32.5 BLEU scores for Hindi-to-Maithili and Maithili-to-Hindi language pairs, respectively. We release the created dataset for further research and development purposes.</p>