CoToGrasp is a generative framework that synthesizes dexterous grasps conditioned on specific contact topologies, trained in an object‑agnostic manner using a feature‑based canonical workspace that decouples functional intent from object geometry. The approach enables zero‑shot generalization to unseen objects and achieves state‑of‑the‑art performance on the DexGraspNet dataset, outperforming existing taxonomy‑guided planners. Physical validation on a robot platform confirms the kinematic feasibility and viability of the synthesized grasps.