When should the agent call a judge instead of writing?

Use it for checks with a fixed answer set such as claim support, page safety, file choice, or patch completion. When the step needs new text, code, or unlistable options, the agent should write it itself.

What keeps a wrong probability from auto-approving work?

The policy maps probabilities to auto for proceed, review for check-another-way, and escalate for stop. The recorded bench had auto-accepted answers correct with misses routed to review instead of through.

How do setup, install, and doctor differ?

Run setup to verify and store the key, install to write agent entries, and doctor for an offline check. Hand-registration needs the uvx path, version-pinned spec with the typesafe extra, and Python 3.12+.

Which agents can it attach to and how is the key protected?

Supported entries include Claude Code, Claude Desktop, Codex CLI and app, Cursor, OpenCode, Pi with its MCP adapter, omp, and Pythinker. Setup never takes the key as an argument so it stays out of shell history.

What does the server need to run?

Python 3.12+, uv, and a POSIX system such as Linux or macOS, plus a TypeSafe API key. The published PyPI package is launched through uvx and speaks MCP over stdio.

What latency and cost should I expect per decision?

In the recorded bench the median round trip was 464.6 ms, and the classify run cost about $0.025 per 1,000 decisions. Those numbers describe that measured run, not a guarantee.

How many tools does it add, and what are they called?

Fifteen judgment tools backed by Jev. Named examples include jev_verify and jev_gate, and in Claude Code they surface under the mcp__jev__ prefix.

CommunityJev Agents & MCP

Jev Judge MCP

MCP server over stdio that gives coding agents fourteen Jev judgment tools, turning evidence plus an enumerable question into auto, review, or escalate actions from calibrated probabilities.

See highlights

Open github.com

Stars: 72
Language: Python
License: MIT
Published by: Community
Author: PyModel
Last verified: Oct 2026

PyModel/jev-judge-mcp

Organization repository on GitHub

View on GitHub

Typed judgment tools for MCP agents. TypeSafe's Jev model as verify, screen, find, classify, rerank, decide, compare, extract, review, gate, and score: the model judges, policy decides auto, review, or escalate.

Forks: 5
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

An MCP server that gives your coding agent fifteen judgment tools backed by TypeSafe's Jev model. The agent hands a tool some evidence and a question it can enumerate: is this claim supported, is this page safe to read, which of these files answers the question, did this patch finish the task. Jev answers with probabilities, usually in under a second (median 464.6 ms round trip in the recorded bench), for about $0.025 per 1,000 decisions on the recorded classify run — and on that benchmark every answer the policy…

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

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification Jev for Ranking Jev for Tool calling

Used inJev for Software & DevTools

Languages: Python

Highlights

Fourteen judgment tools such as verify and gate, with probabilities mapped to auto, review, or escalate.
Four-stage check, find, decide, and ship pipeline aimed at enumerable agent verification steps.
Installer with dry-run, per-agent targeting, removal, checkout pinning, and post-write verification.
Key file under the user config used when the environment variable is unset, with permission warnings.

Watch out

Needs Python 3.12+, uv, a POSIX system, and a verified TypeSafe key; desktop-app installs may need an explicit key flag because they do not inherit shell env. 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

langchain-skill-router

LangChain deepagents middleware that replaces SkillsMiddleware with per-turn judging, loading only the SKILL.md files a turn needs so large catalogs stay out of the prompt.