Hepatocellular carcinoma (HCC) is a common cause of cancer-associated death worldwide. The mitochondrial unfolded protein response (UPR) not only maintains mitochondrial integrity but also regulates cancer progression and drug resistance. However, no study has used the UPR to construct a prognostic signature for HCC. This work aimed to establish a novel signature for predicting patient prognosis, immune cell infiltration, immunotherapy, and chemotherapy response based on UPR-related genes (MRGs). Transcriptional profiles and clinical information were obtained from the TCGA and ICGC databases. Cox regression and LASSO regression analyses were applied to select prognostic genes and develop a risk model. The TIMER algorithm was used to investigate immunocytic infiltration in the high- and low-risk subgroups. Here, two distinct clusters were identified with different prognoses, immune cell infiltration statuses, drug sensitivities, and response to immunotherapy. A risk score consisting of seven MRGs (HSPD1, LONP1, SSBP1, MRPS5, YME1L1, HDAC1 and HDAC2) was developed to accurately and independently predict the prognosis of HCC patients. Additionally, the expression of core MRGs was confirmed by immunohistochemistry (IHC) staining, single-cell RNA sequencing, and spatial transcriptome analyses. Notably, the expression of prognostic MRGs was significantly correlated with sorafenib sensitivity in HCC and markedly downregulated in sorafenib-treated HepG2 and Huh7 cells. Furthermore, the knockdown of LONP1 decreased the proliferation, invasion, and migration of HepG2 cells, suggesting that upregulated LONP1 expression contributed to the malignant behaviors of HCC cells. To our knowledge, this is the first study to investigate the consensus clustering algorithm, prognostic potential, immune microenvironment infiltration and drug sensitivity based on the expression of MRGs in HCC. In summary, the UPR-related classification and prognostic signature could assist in determining the prognosis and personalized therapy of HCC patients from the perspectives of predictive, preventative and personalized medicine.