jev-mcp

MCP server giving any agent Jev's classify, score, yes/no check and allow/confirm/block gating of risky tool calls.

MCP serverbetaMIT

Overview

This jev-mcp is an MCP server that gives agents such as Claude Code, Codex, Cursor and Pi typed decisions from TypeSafe's Jev. It exposes five tools: classify picks one labelled option, score rates input on a scale you define, check returns a yes/no probability, gate risk-screens an action and returns allow, confirm or block, and decide sends many typed questions in one round trip.

The README frames Jev as a fuzzy if inside the agent loop and states typical latency of 70-500 ms and about $0.0004 per decision. It runs straight from GitHub with npx. It works with an official TypeSafe key or with a prepaid key from jevtypesafeai.com, an independent hosted gateway and playground that the README says is not affiliated with TypeSafe; the server picks the endpoint from the key prefix.

What the decision model decides

Type	Question	Answers
Choice	jev_classify: which labelled option fits?	—
Score	jev_score: where does this input sit on your scale (risk, urgency, quality)?	—
Yes / no	jev_check: is this statement true?	yes · no
Choice	jev_gate: should this action run?	allow · confirm · block

How it works

Add the server to an MCP client with a Jev key
The agent calls classify, score, check, gate or decide
Jev returns a typed answer with probabilities
The agent acts, asks for confirmation or blocks based on the result

Strengths

No clone or build; runs via npx from GitHub
Automatic endpoint selection by key type
Version pinning through JEV_MODEL

Caveats

Promotes a third-party paid gateway not affiliated with TypeSafe
Price and latency figures are the README's claims
Not yet published to npm

Getting started

claude mcp add jev -e TYPESAFE_API_KEY=your_key -- npx -y github:codaaiteam/jev-mcp

At a glance

Type: MCP server
Status: beta
Language: JavaScript
Runtime: Node MCP server
Has code: yes
License: MIT

Use cases and integrations

Infra / SDKs / Integrations537

How curated lists describe it

MCP server for Jev (TypeSafe AI's System One model) — give any agent typed, calibrated decisions: classify, score, check, gate risky tool calls. Try free: jevtypesafeai.com
logicrw/awesome-jev-projects

Frequently asked questions

What does jev_gate return?

allow, confirm or block for an action the agent is about to run.

Which keys work?

An official TypeSafe key, or a jv_live_ key from the independent jevtypesafeai.com gateway.

Is jevtypesafeai.com official?

No. The README describes it as an independent playground and guide not affiliated with TypeSafe AI.

Related projects

jev-use
Claude Code, Codex and pi plugin that hands non-writing agent steps to Jev and escalates writing or unsure steps to the LLM.
PluginbetaJavaScript
interlinked-cli
Local guard layer for coding agents that blocks dangerous tool calls by deterministic rules, with optional Jev judgments.
CLIbetaTypeScript
jev-guard (leepokai)
Security hook giving Codex, Copilot, Gemini, Cursor, pi, OpenCode and ACP an auto-mode-like guard via three Jev questions.
LibrarybetaTypeScript
jevwire
Jev decision layer for agents: seven MCP tools, an embeddable library and escalate-only Claude Code hooks at harness boundaries.
MCP serverbetaTypeScript
macos-computer-use-kit
Accessibility-first macOS computer-use MCP server and CLI, with optional Jev guards on targets and inputs before irreversible actions.
PluginbetaPython
master-jev-hook
Local gateway, MCP server and Claude Code hooks that let coding agents delegate explicit choices, classification and scoring to Jev.
MCP serverbetaTypeScript