Integrating expert knowledge into causal structure learning from observational data has proven beneficial. Unlike prior approaches where human knowledge operates independently of structure discovery, we propose a framework for causal discovery with interactive human input. Within our framework, human experts offer edge relations within a local context defined by a machine learning algorithm, iteratively refining the causal graph estimate. Results on simulated and real networks demonstrate superior accuracy with human input. We also analyze the effects of various human inputs and node selection strategies on algorithm performance.