Counterfactual inference asks how an outcome would have differed under unobserved circumstances, e.g., whether a patient would have recovered with a different treatment. Such questions arise whenever decisions depend on the outcomes of alternatives that were not observed.\ \ A common pipeline first estimates the causal graph from observational data and then performs counterfactual inference on the estimated graph. Errors in the graph estimation inevitably propagate to downstream predictions, degrading performance.\ \ To overcome this limitation we introduce the Counterfactual Graph Neural Process (CFGNP), an end‑to‑end model that answers counterfactual queries directly from observational data without first constructing a causal graph. CFGNP leverages Graph Neural Networks (GNNs) to exploit the dual‑graph structure, restricting cross‑world message passing to corresponding factual‑counterfactual variable pairs, thereby preserving essential causal constraints.\ \ Across synthetic and image‑based benchmarks, CFGNP consistently outperforms existing methods. On the German Loan benchmark, the mean squared error (MSE) drops from $4.51 \times 10^{-2}$ to $0.19 \times 10^{-2}$.\ \ Moreover, when trained only on 10‑node synthetic causal graphs, CFGNP retains strong predictive performance on graphs up to 30 nodes without any retraining, demonstrating impressive amortized generalization.\ \ Review: CFGNP eliminates the fragile causal‑graph estimation step, delivering more robust counterfactual predictions and showing strong scalability across graph sizes, making it a valuable tool for real‑world decision‑making.