Antigen tests or polymerase chain reaction (PCR) amplification are currently COVID-19 diagnostic tools. However, developing complementary diagnosis tools is mandatory. Thus, we performed a plasma cytokine array in COVID-19 patients to identify novel diagnostic biomarkers. A discovery–validation study in two independent prospective cohorts was performed. The discovery cohort included 136 COVID-19 and non-COVID-19 patients recruited consecutively from 24 March to 11 April 2020. Forty-five cytokines’ quantification by the MAGPIX system (Luminex Corp., Austin, TX, USA) was performed in plasma samples. The validation cohort included 117 patients recruited consecutively from 15 to 25 April 2020 for validating results by ELISA. COVID-19 patients showed different levels of multiple cytokines compared to non-COVID-19 patients. A single chemokine, IP-10, accurately identified COVID-19 patients who required hospital admission (AUC: 0.962; 95%CI (0.933–0.992); p < 0.001)). The results were validated in an independent cohort by multivariable analysis (OR: 25.573; 95%CI (8.127–80.469); p < 0.001) and AUROC (AUC: 0.900; 95%CI (0.846–0.954); p < 0.001). Moreover, showing IP-10 plasma levels over 173.35 pg/mL identified COVID-19 with higher sensitivity (86.20%) than the first SARS-CoV-2 PCR. Our discover–validation study identified IP-10 as a robust biomarker in clinical practice for COVID-19 diagnosis at hospital. Therefore, IP-10 could be used as a complementary tool in clinical practice, especially in emergency departments.