Untargeted metabolomics using liquid chromatography–mass spectrometry (LC–MS) is currently the gold-standard technique to determine the full chemical diversity in biological samples. However, this approach still has many limitations; notably, the difficulty of accurately estimating the number of unique metabolites profiled among the thousands of MS ion signals arising from chromatograms. Here, we describe a new workflow, MS-CleanR, based on the MS-DIAL/MS-FINDER suite, which tackles feature degeneracy and improves annotation rates. We show that implementation of MS-CleanR reduces the number of signals by nearly 80% while retaining 95% of unique metabolite features. Moreover, the annotation results from MS-FINDER can be ranked according to the database chosen by the user, which enhance identification accuracy. Application of MS-CleanR to the analysis of Arabidopsis thaliana grown in three different conditions fostered class separation resulting from multivariate data analysis and led to annotation of 75% of the final features. The full workflow was applied to metabolomic profiles from three strains of the leguminous plant Medicago truncatula that have different susceptibilities to the oomycete pathogen Aphanomyces euteiches. A group of glycosylated triterpenoids overrepresented in resistant lines were identified as candidate compounds conferring pathogen resistance. MS-CleanR is implemented through a Shiny interface for intuitive use by end-users (available at https://github.com/eMetaboHUB/MS-CleanR).