MCP

Jev Review: A Continuous Software Quality Review MCP Plugin for AI Coding Agents

Jev Review is best suited for having AI coding agents continuously review code quality during development with 1–10 scores, but you must first obtain a Jev API key, and the task, diff, and file contents you provide are sent to TypeSafe's cloud API.

Official source · NiazMorshed2007/jev-review

Install

npx plugins add NiazMorshed2007/jev-review

Jev Review is a local-first MCP plugin driven by Jev, suited for developers who want their AI coding agents to continuously perform software quality reviews during development: it has no hosted backend, database, or telemetry, but the prerequisite is that you must first obtain a Jev API key, and every review's content (task, diff, file contents, repository context) is sent via TLS directly to TypeSafe's cloud API — so it is not suitable if you are unwilling to send code snippets to an external service.

Who It's For

  • Developers already using AI coding agents (such as Claude Code, Codex, Cursor, OpenCode) who want repeated code quality scores and improvement suggestions throughout their task workflow.
  • Scenarios where review results need to be comparable and trackable: passing in previousEvaluation yields per-metric score changes (deltas).
  • Not suitable for users without a Jev API key — configuring the key is a hard prerequisite; the evidence does not state whether there is a free tier or trial option, and this remains unknown.
  • Not suitable for users who want fully offline reviews: although the process starts locally, review requests are sent to https://api.typesafe.ai/v1/systemone.

First Attempt in 3 Minutes

Prerequisites: Node.js 20 or higher, plus a Jev API key.

  1. Export the key (the official approach; replace "your-key" with your real key):
export JEV_API_KEY="your-key"
  1. Install the plugin (the official main path; installs from GitHub, no npm publishing needed):
npx plugins add NiazMorshed2007/jev-review
  1. Call the single tool jev_review in your client for a low-risk review: pass in task (the task description) and either diff or files (path/content). Note that at least one current-context field must be provided; the tool never reads the repository automatically — everything is supplied explicitly by you.

Success signal: the response contains independent 1–10 scores for each applicable metric and a 0–1 confidence value, along with rubric hints for the weak dimensions.

Privacy notes: this tool has no read/write distinction — its very purpose is to send the task, diff, file contents, and repository context you provide in the parameters to the remote Jev API. The key is only placed in the TLS Authorization header, and the official documentation states that the key is not stored or logged, nor are repository files automatically discovered or uploaded. Before your first attempt, confirm that the code snippets you will send are safe to transmit.

Use Cases

Case 1: Quick pre-commit review of a change. Scenario: you have just finished a feature diff and want a quality score before making the formal commit. Call jev_review, filling in the task description in task and the diff in the diff parameter. Expected result: 1–10 scores and 0–1 confidence for the applicable metrics, plus rubric hints pointing out weak dimensions, helping you decide whether to keep iterating before committing.

Case 2: Continuous quality tracking across iterations of a task. Scenario: you have revised your code based on the previous round of review feedback and want to know whether the scores improved. When calling jev_review, in addition to this round's task/diff/files, pass in the previous round's evaluation as previousEvaluation. Expected result: the response contains per-metric score changes (deltas), which you can use to judge whether your changes were effective. The evidence does not show the exact delta output format; details beyond this point are subject to the actual response.

Usage Boundaries

  • Read/write nature: jev_review is the only tool; it does not modify local files or automatically scan the repository. But it is a remote call — the content you provide is sent to TypeSafe's Jev API. The key travels via the TLS Authorization header, officially stated as neither stored nor logged, and repository files are never uploaded automatically.
  • Credential requirement: the JEV_API_KEY environment variable must be set.
  • Parameter requirement: at least one current-context field among task, diff, files, repositoryContext, and previousEvaluation is required; otherwise the call is meaningless.
  • Side effects and confirmation points: the only side effect is data being sent out; there is no local state write or repository change. If your organization does not allow code to leave the premises, confirm policy first before using it.
  • Official limitations: supported clients are Claude Code, Codex, Cursor, and OpenCode, all of which launch the bundled dist/server.js process locally in the same way (stdio); behavior is identical across all clients.
  • Not yet confirmed: the specific list of scoring metrics, API quotas and pricing, and the process for obtaining a key are all not stated in the evidence — refer to official channels.