Stroke, the second leading cause of death and disability, causes massive cell death in the brain followed by secondary inflammatory injury initiated by disease associated molecular patterns released from dead cells. Nonetheless, the evidence regarding the causal relationship between inflammatory cytokines and stroke subtypes is obscure. To leverage large scale genetic association data to investigate the interplay between circulating cytokines and stroke, we adopted a two-sample bi-directional Mendelian randomization (MR) analysis. Firstly, we performed a forward MR analysis to examine the associations of genetically determined 31 cytokines with 6 stroke subtypes. Secondly, we conducted a reverse MR analysis to check the associations of 6 stroke subtypes with 31 cytokines. In the forward MR analysis, genetic evidence suggests that 21 cytokines were significantly associated with certain stroke subtype risk with |β| ranging from 1.90 × 10
−4
to 0.74. In the reverse MR analysis, our results found that five stroke subtypes (intracerebral hemorrhage (ICH), large artery atherosclerosis ischemic stroke (LAAS), lacunar stroke (LS), cardioembolic ischemic stroke (CEI), small-vessel ischemic stroke (SV)) caused significantly changes in 16 cytokines with |β| ranging from 1.08 × 10
−4
to 0.69. In particular, those five stroke subtypes were statistically significantly associated with C-reactive protein (CRP). In addition, ICH, LAAS, LS and SV were significantly correlated with vascular endothelial growth factor (VEGF), while LAAS, LS, CEI and SV were significantly related to fibroblast growth factor (FGF). Moreover, integrated bi-directional MR analysis, these factors (IL-3Rα, IL-6R, IL-6Rα, IL-1Ra, insulin-like growth factor-1(IGF-1), IL-12Rβ2) can be used as predictors of some specific stroke subtypes. As well as, IL-16 and C–C motif chemokine receptor 7 (CCR7) can be used as prognostic factors of stroke. Our findings prognostic identify potential pharmacological opportunities, including perturbation of circulating cytokines for both predicting stroke risk and post stroke treatment effects. As we conducted a comprehensive search and analysis of stroke subtype and cytokines in the existing publicly available GWAS database, the results have good population-generalizability.