Peptides, with their unique physicochemical properties, are increasingly recognized as powerful therapeutic molecules for pharmaceutical development. Knottins are a unique subclass of disulfide-constrained peptides, characterized by a complex, knot-like topology formed by six connected cysteine residues. This topology provides valuable intrinsic properties, such as exceptional thermal and enzymatic stability. Consequently, knottins have emerged as promising scaffolds for developing large phage display libraries for hit discovery. However, traditional library design approaches often introduce random mutations without sufficient effort on preserving the scaffold's crucial topology, leading to compromised stability and challenging structural characterization. Here, we leverage Protein MPNN to computationally design novel sequences for EETI-II, a well-studied knottin peptide, with an emphasis on conserving its native topology. Through an integrated approach combining computational design with biochemical foldability assays and structural modeling, we validated the ability of Protein MPNN to generate diverse, foldable, and stable EETI-II variants. These computational insights then guided the design of large, diverse phage display libraries, which successfully yielded peptide binders with maintained structural integrity. Overall, this multidisciplinary approach establishes a robust framework for the rational design and discovery of stable, functional knottin peptides, expanding the therapeutic potential of this molecular class.