jev-judge-mcp
MCP server giving coding agents typed judgment tools (verify, screen, rerank, gate, score) backed by Jev, with auto/review/escalate policy.
Overview
jev-judge-mcp is an MCP server that gives a coding agent judgement tools backed by Jev, for checks whose answers can be enumerated: is this claim supported, is this page safe to read, which of these files answers the question, did this patch finish the task. List summaries name tools such as verify, screen, find, classify, rerank, decide, compare, extract, review, gate and score; the README mentions fifteen. Policy turns probabilities into auto (proceed), review (check another way) or escalate (stop).
The README reports a recorded bench with a median 464.6 ms round trip and about $0.025 per 1,000 decisions on a classify run, where every auto-accepted answer was correct and misses were routed to review. It advises the agent to write new text or code itself when the step is not enumerable. Installation uses uvx with setup, install and offline doctor commands that configure agents to launch a pinned PyPI package. It needs Python 3.12+, uv and a POSIX system.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this claim supported by the evidence? | — |
| Yes / no | Is this page safe to read? | — |
| Choice | Which of these files answers the question? | — |
| Yes / no | Did this patch finish the task? | — |
How it works
- Agent calls a judgement tool with evidence and an enumerable question
- Jev returns probabilities
- Policy maps to auto, review or escalate
- Agent proceeds, double-checks or stops
Strengths
- Many ready-made judgement tools
- Policy routes misses to review
- Pinned, reproducible install
Caveats
- Benchmark is a recorded run, not a broad evaluation
- POSIX only
Getting started
`uvx --from 'jev-judge-mcp[typesafe]' jev-judge-mcp setup`, then `... install` and `... doctor`.At a glance
- Type
- MCP server
- Status
- beta
- Language
- Python
- Runtime
- MCP server (uv)
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
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.
logicrw/awesome-jev-projectsTyped judgment tools for MCP agents. TypeSafe.
wh000wh000/awesome-jev-liveTyped 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.
hellogumbo/awesome-jevMCP server that gives a coding agent eleven judgment tools backed by Jev, for the checks whose answers can be enumerated.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
Answers to enumerable checks such as claim support, page safety, file relevance and task completion.
What happens on low confidence?
Policy routes to review or escalate rather than auto.
What does it cost?
About $0.025 per 1,000 decisions on the recorded classify run.
Related projects
Small stdio MCP server with ask, rank and verify tools backed by Jev, a local event log and installers for several agent harnesses.
MCP serverbetaTypeScriptEvidence gate for RAG: a System One model judges whether retrieved excerpts support an answer before the LLM is allowed to speak.
MCP serverbetaPythonpip-installable Jev kit: a CLI for verify, screen, classify, extract, route and rerank, plus an MCP server for the primitives.
MCP serverbetaPythonSelf-hosted Jev MCP gateway on Cloudflare Workers with nine tools for intent routing, reranking and batch judgments.
MCP serverbetaTypeScriptAgent skill that teaches coding agents to hand bulk semantic judgments to Jev and to find where a product could use Jev decisions.
Agent skillbetaMCP server and agent skill for designing Jev judgments: question linting, batch evaluation, threshold calibration and reranking.
MCP serverbetaTypeScript