Skip to content

jev-mcp-spring

Java/Spring Boot MCP server exposing Jev classify, score, check and patch-gate tools over HTTP instead of stdio.

MCP serverbetaApache-2.0

Overview

Jev MCP Server (Spring) is a Java and Spring Boot MCP server that exposes TypeSafe Jev as typed-judgment tools over HTTP (Streamable or SSE). It targets MCP clients that reach a running service rather than spawning a local subprocess, and the Java side of the ecosystem. It wires the jev-spring-boot-starter client into Spring AI's MCP server support.

Tools are jev_classify (label, confidence, probabilities), jev_score against an ordered rubric, jev_check returning a probability with a flag, uncertain or clear band, jev_gate that verifies completion claims against a diff and evidence and scores merge risk to give pass or review, and jev_health. Failures surface at the MCP protocol level with a short safe message and never echo response bodies or stack traces. A generic ask tool, candidate matching and prompt-injection screening are listed as not implemented yet. The author notes it is an independent community project.

What the decision model decides

TypeQuestionAnswers
Choicejev_classify: which given label fits the state?—
Yes / nojev_check: does this yes/no condition hold?flag · uncertain · clear
Choicejev_gate: should this patch merge given its claims and evidence?pass · review

How it works

  1. Run the Spring Boot service with TYPESAFE_API_KEY
  2. Point an HTTP-capable MCP client at it
  3. The agent calls classify, score, check or gate
  4. Typed results return; failures surface as isError with safe messages

Strengths

  • HTTP transport for clients that cannot spawn subprocesses
  • Safe error handling that never leaks bodies or stack traces
  • Apache-2.0 licence and CI

Caveats

  • Implements a subset of tools; ask, match and screen are not yet available
  • Requires Java 17+ and your own TypeSafe key
  • Independent, unofficial project

Getting started

With Java 17+, export TYPESAFE_API_KEY and start the Spring Boot application as described in the README.

At a glance

Type
MCP server
Status
beta
Language
Java
Runtime
Spring Boot MCP server (HTTP/SSE)
Has code
yes
Has tests
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • On success each tool returns its typed result directly. On failure the tool call fails at the MCP protocol level (\`isError: true\`) with a short, safe message — no response body or stack trace is ever echoed back.

    logicrw/awesome-jev-projects

Frequently asked questions

Why HTTP instead of stdio?

It is for MCP clients that connect to a running service rather than spawning a local process.

What does jev_gate decide?

It verifies completion claims against a diff and evidence, scores merge risk and returns pass or review.

Which tools are missing?

A generic ask tool, candidate matching and prompt-injection screening are tracked as follow-ups.

Related projects

  • Agent Zero plugin adding a typesafe_query tool that asks Jev Choice, Noul and Score questions and shows probability cards in chat.

    ProjectbetaPython
  • CLI, MCP server and Claude Code plugin for closed-form yes/no, choice, score and rank decisions via Jev, OpenAI Decisions or Claude.

    MCP serverbetaTypeScript
  • Hermes Agent plugin exposing Jev as four coding-workflow tools: jev_check, jev_route, jev_score and jev_evaluate.

    PluginbetaPython
  • Gives Claude Code, Codex, Pi and OpenCode five typed Jev tools (classify, check, score, rank, ask) plus a skill on when to use them.

    MCP serverbetaTypeScript
  • MCP server exposing Jev Noul, Choice and Score via OpenRouter so coding agents can make quick decisions without reasoning tokens.

    MCP serverbetaTypeScript
  • Zero-dependency Node bridge that exposes Jev through an OpenAI-compatible API, plus a CLI, code-audit tool and auto-fix loop.

    CLIbetaJavaScript