Emotion-Cause Pair Extraction (ECPE) task, which aims at identifying and extract emotion clauses and corresponding cause clauses. Existing approaches typically employ sequential encoding of features in a predetermined order, which results in imbalanced feature interactions between tasks, whereby information can only flow from the emotion/cause clause encoder to the pair encoder. Additionally, the approach is not sensitive to long-distance emotion-cause pairs, and the relatively low precision of the extracted ground for cause clauses. To address these issues, this paper proposes a method for Emotion-Cause Pair Extraction based on the Machine Reading Comprehension (MRC) framework with Joint Coding (MRCJE). This method improves the accuracy of auxiliary tasks such as emotion extraction and cause extraction by concatenating queries and clause displays. It also uses an undirected isomorphic graph to transfer information between clauses and pairs, and generates both pairs and clause features to model causal relationships in clauses, balancing the information flow between emotion clauses, cause clauses and pairs. The method was experimentally demonstrated on a Chinese benchmark corpus, and the results demonstrated that it achieved better results than the baseline model.