Jev Review

Jev Review is a local MCP server (stdio, one focused tool: jev_review) for continuous software-quality checkpoints while an AI coding agent works. Intended rhythm: after a coherent implementation slice, after a score-driven improvement, and before final handoff — establish a baseline, inspect, hypothesize about weak dimensions, improve, re-score. Jev returns typed decisions (Score, Choice, Noul) rather than a prose essay; the server validates and converts them into metric scores, confidence levels, coarse rubric hints, and comparisons with the previous evaluation, while the coding agent stays responsible for diagnosing causes and changing code. Fifteen dimensions are always evaluated when context suffices — correctness and requirement fit, cognitive complexity, readability and intent, modularity and cohesion, coupling and dependency quality, changeability and change amplification, abstraction and API design, project and file structure, duplication and reuse, maintainability, testability and test quality, reliability and error handling, security, consistency and conventions, documentation and explainability — plus four evidence-gated ones: performance and resource efficiency, scalability and flexibility, compatibility and API stability, observability and operability. There is no hosted backend, database, telemetry or author-operated proxy: the only remote request goes directly to the configured Jev API. Node.js 20+, distributed from GitHub (no npm package), MIT.