We propose DCCD-CONF, a scalable, theoretically grounded framework for learning nonlinear cyclic causal graphs with unmeasured confounders using interventional data.