mcp-server-jev
Community MCP server giving Codex, Claude and other clients two tools for typed Jev decisions: evaluate and classify.
Overview
Jev MCP Server gives Codex, Claude and other MCP clients a tool for typed decisions with Jev: classify documents, triage tickets, score relevance or review content against a rubric, asking several independent questions in one call and getting structured answers instead of prose. It exposes two tools: jev_evaluate for mixed noul, choice and score questions, and jev_classify for repeated classification against the same predefined classes. Responses include the actual model, token usage, latency and attempt count, using TypeSafe's native API shapes.
The README is explicit about limits and good practice: Jev evaluates only the content supplied; it does not fetch URLs, search, write explanations or act, and confidence describes a distribution rather than proving accuracy. Agents should retrieve evidence first, give Jev original text and concrete class definitions, include a no-match class, and not send their own drafted verdict as evidence. Eight reproducible live scenarios, covering support, search intent, lead qualification, editorial checks, duplicates, adversarial messages, rubric design and batching, are published with failures and raw results. A French installation guide and article are provided.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Any yes/no judgment about supplied content | — |
| Choice | Which predefined class fits this item? (jev_classify) | — |
| Score | How does this content rate against the rubric? | — |
How it works
- Install the MCP server and set the API key privately
- Agent retrieves evidence and defines classes or rubric
- Agent calls jev_evaluate or jev_classify
- Typed answers return with usage and latency
- Agent or human keeps responsibility for consequential decisions
Strengths
- Simple two-tool interface
- Published live scenarios including failures
- Clear guidance on proper use
- French documentation
Caveats
- Community project, not affiliated with TypeSafe
- Requires a TypeSafe key
- Does not fetch or verify facts
Getting started
Follow docs/installation-fr.md or the README to install and register the server with Codex or Claude.At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- MCP server
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Typed AI decisions for Codex, Claude and any MCP client, powered by TypeSafe Jev. Classify, score and evaluate with one generic tool.
logicrw/awesome-jev-projects
Frequently asked questions
What tools does it expose?
jev_evaluate and jev_classify.
Can Jev look things up?
No, it only evaluates the content you supply.
Are results tested?
Eight live scenarios are published with raw results, including failures.
Related projects
CLI, MCP server and Claude Code plugin for closed-form yes/no, choice, score and rank decisions via Jev, OpenAI Decisions or Claude.
MCP serverbetaTypeScriptMCP server exposing Jev as one evaluate tool, plus a harness comparing Jev's judgments with general LLMs on shared datasets.
MCP serverbetaJavaScriptClaude Code and Codex skills plus a study course for building with Jev, including Jevaluate, which rates how well projects use Jev.
ProjectbetaSkill and plugin for Codex, Claude Code and other agents that helps them use Jev for bounded judgments and measure the effect.
Agent skillbetaJavaScriptDecision plugin for agent harnesses where Jev recommends tools, skills or owners and scores output, for DeepSeek Harness and iPolloWork.
Agent skillbetaTypeScriptEval-first MCP server for Jev: prototype questions, map them over many items, and measure accuracy and thresholds on labelled data.
MCP serverbetaTypeScript