Benchmarks & research

jev-mcp

@arunav25★ 5JavaScriptMITupdated 2026-09-17▶ runnable

Connect JEV to MCP clients and compare its judgments against general-purpose LLMs using shared datasets and measurable accuracy.

arunav25/jev-mcp ↗

Where it calls Jev

export const DEFAULT_BASE_URL = "https://api.typesafe.ai";

src/config.js:6

The link points at the commit we read, so the line number still holds.

What it asks Jev

These question sets are lifted from this project's source exactly as written.

from test/tool.test.js:60

  1. urgenttrue/false

    urgency?

Run it on your own text

The code
import { experimental_evaluate as evaluate } from 'ai';

const { answers } = await evaluate({
  model: 'typesafe-ai/jev',
  state,
  questions: {
    urgent: { type: 'boolean', instructions: 'urgency?' },
  },
});

Other projects in this category