decisions-judge-mcp

MCP server exposing a judge tool for typed noul, choice and score questions over JSON state, backed by Jev.

MCP serverbeta

Overview

decisions-judge-mcp gives AI agents typed decisions as an MCP tool. The judge tool asks yes/no probability (noul), choice among options or score on ordered levels about any JSON application state, with all questions answered in one request. It is backed by TypeSafe's Jev through the official @typesafe-ai/sdk, with an opt-in Cloudflare Workers AI provider.

It runs over stdio by default via npx and works with Claude Desktop, Codex, Goose and Pi using the same configuration shape. An opt-in stateless Streamable HTTP transport serves POST /mcp with a health endpoint; the README warns that this endpoint is unauthenticated and must sit behind an OAuth proxy in production. The tool returns typed answers code can branch on, with a guaranteed fallback envelope on failure.

What the decision model decides

Type	Question	Answers
Yes / no	Any yes/no question about the supplied JSON state(inferred)	yes · no
Choice	Any choice among supplied options(inferred)	—
Score	Any rating on supplied ordered levels(inferred)	—

How it works

Configure the MCP server with TYPESAFE_API_KEY
Agent calls judge with state and questions
Server forwards one request to Jev (or Workers AI)
Agent branches on typed answers or the fallback envelope

Strengths

One request for several typed questions
Works across several MCP clients
Fallback envelope on errors

Caveats

HTTP transport is unauthenticated by default
Needs a TypeSafe or Cloudflare key

Getting started

Add an MCP server running npx -y decisions-judge-mcp with TYPESAFE_API_KEY in env.

At a glance

Type: MCP server
Status: beta
Language: TypeScript
Runtime: MCP server (stdio or HTTP)
Has code: yes
Has tests: yes
Stars: 5

Use cases and integrations

Infra / SDKs / Integrations537

How curated lists describe it

Typed decisions for AI agents as an MCP tool: yes/no probability (noul), choice, and score in one fast request. Backed by the TypeSafe System One model.
wh000wh000/awesome-jev-live
Typed decisions for AI agents as an MCP tool: yes/no probability (noul), choice, and score in one fast request. Backed by the TypeSafe System One model.
hellogumbo/awesome-jev

Frequently asked questions

What tool does it add?

A judge tool for noul, choice and score questions about JSON state.

Which providers?

Jev via the TypeSafe SDK, or opt-in Cloudflare Workers AI.

Can I run it remotely?

Yes with JUDGE_TRANSPORT=http, but put an authenticating proxy in front.

Related projects

Jev DSH Decision Engine
Decision plugin for agent harnesses where Jev recommends tools, skills or owners and scores output, for DeepSeek Harness and iPolloWork.
Agent skillbetaTypeScript
jev-mcp (arunav25)
MCP server exposing Jev as one evaluate tool, plus a harness comparing Jev's judgments with general LLMs on shared datasets.
MCP serverbetaJavaScript
jev-mode (tiffygk)
Claude Code and Codex skills plus a study course for building with Jev, including Jevaluate, which rates how well projects use Jev.
Projectbeta
jevriel
Skill and plugin for Codex, Claude Code and other agents that helps them use Jev for bounded judgments and measure the effect.
Agent skillbetaJavaScript
JevX
CLI and MCP tool that finds hard-coded judgment calls in your code, scores how well Jev fits, and replaces strong fits with a fallback.
ProjectbetaTypeScript
mcp-server-jev
Community MCP server giving Codex, Claude and other clients two tools for typed Jev decisions: evaluate and classify.
MCP serverbetaTypeScript