JevReview
jev-review is a local Model Context Protocol (MCP) plugin that gives AI coding agents a structured software-quality score while they work. It runs as a local Node.js process with no hosted backend, sending requests directly from the user's machine to the Jev API using the user's own API key.
The tool addresses a specific gap in agent-driven coding: an agent implementing a change has no independent signal on whether the result is actually good, only whether it compiles or passes tests. jev-review closes that loop by scoring a focused diff across independent quality dimensions -- correctness, complexity, changeability, modularity, tests, and security -- and returning typed results rather than a free-form review essay. The intended workflow is a checkpoint pattern: the agent implements a slice, calls the tool for a baseline score, forms a hypothesis about the weakest dimension, improves the code, and rescores before moving on.
It shares its scoring engine, Jev, with the previously absorbed JevUltrafast blip, which applies the same engine to browser automation rather than code review. That shared lineage suggests the underlying scoring approach is being productized across multiple agent-tooling surfaces rather than being a one-off experiment.
jev-review sits in Assess. It shows strong early interest for its age (196 stars within days of a 2026-09-17 creation date), supports four major coding agent clients (Claude Code, Codex, Cursor, OpenCode), and keeps data local by design. But it has no first-person production use behind it, and the accuracy of its quality scores is unverified against real-world review outcomes.
