The main obstacles for practical uses of cytosol-penetrating peptides and proteins include their lack of cell- or tissue-specific targeting and limited cytosolic access owing to the poor endosomal escape ability. We have previously reported a cytosol-penetrating, human IgG1 antibody TMab4-WYW, generally referred to as a cytotransmab (CT), which reaches the cytosol of living cells but nonspecifically because it is endocytosed via a ubiquitously expressed receptor called heparan sulfate proteoglycan (HSPG). Here, our aim was to construct a next-generation CT with tumor cell specificity and improved endosomal escape efficiency. We first substantially reduced the HSPG-binding activity of TMab4-WYW and then fused a cyclic peptide specifically recognizing tumor-associated epithelial cell adhesion molecule (EpCAM) to the N terminus of the light chain for EpCAM-mediated endocytosis, while maintaining the endosomal escape ability in the light chain variable domain (VL), thus generating epCT05. Then, we separately engineered another CT, dubbed epCT65-AAA, with an endosomal escape ability only in the heavy chain variable domain (VH) but not in VL, by functional grafting of the endosomal escape motif of epCT05 VL to the VH. We finally combined the heavy chain of epCT65-AAA and the light chain of epCT05 to create epCT65 with endosomal escape capacity in both the VH and VL. epCT65 effectively localized to the cytosol of only EpCAM-expressing tumor cells and showed approximately twofold improved endosomal escape efficiency, as compared with CTs with endosomal escape motifs in either VH or VL. The full-IgG format CT, epCT65, with a tumor cell-specific cytosol-penetrating activity, has a great potential for practical medical applications, e.g., as a carrier for cytosolic delivery of payloads.