Tuberculous meningitis (TBM)—the extrapulmonary form of tuberculosis, is the most severe complication associated with tuberculosis, particularly in infants and children. The gold standard for the diagnosis of TBM requires cerebrospinal fluid (CSF) through lumbar puncture—an invasive sample collection method, and currently available CSF assays are often not sufficient for a definitive TBM diagnosis. Urine is metabolite-rich and relatively unexplored in terms of its potential to diagnose neuroinfectious diseases. We used an untargeted proton magnetic resonance (1H-NMR) metabolomics approach to compare the urine from 32 patients with TBM (stratified into stages 1, 2 and 3) against that from 39 controls in a South African paediatric cohort. Significant spectral bins had to satisfy three of our four strict cut-off quantitative statistical criteria. Five significant biological metabolites were identified—1-methylnicotinamide, 3-hydroxyisovaleric acid, 5-aminolevulinic acid, N-acetylglutamine and methanol—which had no correlation with medication metabolites. ROC analysis revealed that methanol lacked diagnostic sensitivity, but the other four metabolites showed good diagnostic potential. Furthermore, we compared mild (stage 1) TBM and severe (stages 2 and 3) TBM, and our multivariate metabolic model could successfully classify severe but not mild TBM. Our results show that urine can potentially be used to diagnose severe TBM.