Does registering the MCP server force my agent to use it?

No. Installing the server only adds the tool; the standing rule in examples/agent-instructions.md must go into CLAUDE.md, Cursor rules, or AGENTS.md so every finished task calls submit_for_review.

What happens when the request states nothing checkable?

It returns needs_clarification with the reason and what to add, consumes no retry, and the same submission is judged once acceptance criteria arrive in notes.

Which outcomes burn through the retry budget?

Approved and needs_fixes consume the server-side budget of 10 by default; review_unavailable and needs_clarification are not verdicts on the change and consume nothing.

How does it know my project's naming conventions?

It samples the repo's own declarations for naming, indent, and quoting, then cites file-and-line evidence such as camelCase identifiers in a snake_case repo.

How do I install and register it?

Clone the repository, run npm install and npm run build, then add it to your client (for Claude Code: claude mcp add vibecheck --scope user -- node /absolute/path/to/dist/index.js) or a project .mcp.json, passing TYPESAFE_API_KEY. Without a key it falls back to an offline heuristic judge that labels every verdict as not a real review.

What inputs does submit_for_review take and what comes back?

It takes the original task_description verbatim, an optional project_root, changed_files, raw test_results and notes; failing tests are a hard gate. It returns a verdict, per-dimension scores, an ordered feedback list with file-and-line evidence, attempts_remaining and a next_action the agent can act on directly.

What are the five verdicts?

approved, needs_fixes, max_retries_exceeded, review_unavailable when the judge cannot be reached, and needs_clarification when the request states nothing checkable. The last two are not verdicts on the change and consume no retry; at the retry limit the agent is told to stop and report outstanding issues rather than loop.

CommunityJev Agents & MCP

Vibecheck MCP

MCP done-gate for coding agents: it diffs git reality, measures conventions deterministically, asks Jev eight quality questions, and returns a fix list.

See highlights

Open github.com

Stars: 0
Language: TypeScript
License: MIT
Published by: Community
Author: BilalUchiha
Last verified: Oct 2026

BilalUchiha/vibecheck-mcp

User repository on GitHub

View on GitHub

An MCP server that gives AI coding agents an external "done" criterion — judged by TypeSafe AI's Jev decision model. Agents submit work before declaring completion and get specific, evidence-backed fix lists instead of a vague pass/fail.

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

An external "done" gate for AI coding agents — judged by TypeSafe AI's Jev. Your agent thinks it's finished. vibecheck disagrees, lists exactly what to fix, and only then lets it say the word "done".

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Tool calling

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Git-authoritative change set plus deterministic checks for casing, hardcodes, failures, TODOs, and weak tests.
One Jev call asks 8 scoring plus 8 diagnostic questions; code owns thresholds and hard gates.
Verdicts include approved, needs_fixes, retries-exceeded, unavailable, and needs-clarification states.
Project config, presets, per-dimension thresholds, scope, and review log via three MCP tools.

Watch out

The offline heuristic fallback is for plumbing only and labels every verdict as not a real review; live gates need a Jev key. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

promptrejectormcp

Prompt-injection scanner that combines deterministic checks with focused Jev judgments and a configurable reasoning model, exposed over CLI, SDK, HTTPS, MCP, and editor plugins.

Jev Agents & MCP#community#github

Community

revsmoke/promptrejectormcpTypeScript

opencode-smart-reasoning

OpenCode plugin that asks Jev once per prompt how much reasoning is needed, then maps minimal to xhigh effort onto the model's own variants without rewriting prompts.

Jev Agents & MCP#community#github

Community

d0nj/opencode-smart-reasoningTypeScript

pi-jev-gate

Pi extension that checks file-mutating tool calls against written rules in constraints.md, asking Jev which rule a change violates and offering rework with the rule text.

Jev Agents & MCP#community#github

Community

hcl-z/pi-jev-gateTypeScript

jev-mcp

Community Node.js MCP server exposing Jev as four read-only advisory tools for bounded probabilistic decisions, with the host model keeping final authority over every action.

Jev Agents & MCP#community#github

Community

Afloat16/jev-mcpTypeScript

agentconnect

1.5k

Open-source platform that puts ACP-compatible agents such as Claude Code, Codex, and Pi inside the team's Slack, Telegram, Discord, and GitHub workflows, with reusable Jev decisions for routing, response gating, and reviewer selection.

Jev Agents & MCP#community#jevbest

Community

agentconnect-md/agentconnectTypeScript

488

A coding agent built on pi that hands its non-code decisions — what enters the context, whether a command is safe, whether the work is done — to a small judge model at 35 decision points per turn.

Jev Agents & MCP#community#jevbest

Community

qybaihe/muTypeScript

Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai