An intelligent code review system and method based on MCP protocol

Abstract

The application provides an intelligent code review system and method based on an MCP protocol, which receives code hosting platform events, constructs event fingerprints based on standardized difference summaries, and realizes idempotent arrangement in combination with lease tokens; context self-adaptive construction is performed through risk scoring, cross-file influence propagation scoring, and budget constraints; context is obtained on demand under the same snapshot based on a task-level minimum capability tool list; and fusion review, evidence package output, and feedback self-learning are performed in combination with a rule engine and a large model, so as to improve code review efficiency, accuracy, and interpretability. Through event idempotent arrangement, risk self-adaptive context scheduling, MCP fine-grained on-demand data acquisition, multi-channel fusion verification, and feedback self-learning mechanisms, the application can reduce the overhead of repeated review and invalid context, improve the accuracy, real-time performance, interpretability, and engineering operation stability of code review.

Images