Immune checkpoint blockade using monoclonal antibodies (mAbs) has revolutionized cancer treatment. However, a large fraction of patients do not respond, and systemic administration of mAbs can generate toxicity. Here, we aimed to improve this therapy by optimizing both the therapeutic antibodies and their delivery route. We first generated novel nanobodies against PD-1 and PD-L1 capable of inhibiting both human and mouse interactions. Nanobodies have interesting properties such as a small size, high tissue penetration and high stability. Dimerization of nanobodies by fusion to the Fc domain of mouse IgG reduced significantly the IC50 for PD-1/PD-L1 interaction compared to their monomeric versions. To deliver nanobodies into tumors, which could increase local bioavailability and decrease toxicity, we used a self-amplifying RNA vector based on Semliki Forest Virus (SFV). SFV provides high and transient expression of the gene of interest, induces immunogenic cell death and type I interferon responses. SFV viral particles expressing dimeric nanobodies showed a potent antitumor activity in MC38 and B16 tumor models, outperforming SFV vectors expressing mAbs against PD-1 and PD-L1 and inducing strong inflammation in tumors without toxicity. The SFV system can also be employed as a non-viral vector, which could be easier to translate to the clinic. To test this possibility, we used a DNA plasmid encoding the SFV-anti-PD-1 nanobody vector and delivered it to MC38 tumors by electroporation, yielding a strong antitumor effect. SFV vectors are attractive tools that could improve the response rate to PD-1/PD-L1 inhibitors, avoiding subjecting the patients to high doses of antibodies.