Cystine-dense peptides (CDPs) are a miniprotein class that can drug difficult targets with high affinity and low immunogenicity. Tools for their design, however, are not as developed as those for small-molecule and antibody drugs. CDPs have diverse taxonomic origins, but structural characterization is lacking. Here, we adapted Iterative Threading ASSEmbly Refinement (I-TASSER) and Rosetta protein modeling software for structural prediction of 4298 CDP scaffolds and performed in silico prescreening for CDP binders to targets of interest. Mammalian display screening of a library of docking-enriched, methionine and tyrosine scanned (DEMYS) CDPs against PD-L1 yielded binders from four distinct CDP scaffolds. One was affinity-matured, and cocrystallography yielded a high-affinity (KD = 202 pM) PD-L1-binding CDP that competes with PD-1 for PD-L1 binding. Its subsequent incorporation into a CD3-binding bispecific T cell engager produced a molecule with pM-range in vitro T cell killing potency and which substantially extends survival in two different xenograft tumor-bearing mouse models. Both in vitro and in vivo, the CDP-incorporating bispecific molecule outperformed a comparator antibody-based molecule. This CDP modeling and DEMYS technique can accelerate CDP therapeutic development.