Jev MCP server
jev-mcp is the open MCP server that exposes Jev's decision primitives — classify, score, check, gate, decide — as tools any MCP client (Claude Code, Cursor, Codex, OpenCode) can call. It runs straight from GitHub with npx and picks the right endpoint from your key.
Every example needs a jv_live_ key. Grab one on the pricing page, export it as JEV_API_KEY, and you're ready. Get a key →
If you reached this page because a Jev MCP setup returned an authentication error, the fix is almost always the endpoint — see the gotcha below. If you're setting it up fresh, it's two lines.
Install
The fastest path in Claude Code:
claude mcp add jev -e TYPESAFE_API_KEY=your_key -- npx -y github:codaaiteam/jev-mcpOr add it to any MCP client's config (Cursor, Codex, OpenCode, your own):
{
"mcpServers": {
"jev": {
"command": "npx",
"args": ["-y", "github:codaaiteam/jev-mcp"],
"env": { "TYPESAFE_API_KEY": "jv_live_..." }
}
}
}The five tools
- jev_classify — pick one of your labelled options (routing, categorization, intent)
- jev_score — rate the input on an ordered scale you define (risk, urgency, quality)
- jev_check — a calibrated yes/no probability (gates, filters, guardrails)
- jev_gate — risk-screen an action before it runs → allow / confirm / block
- jev_decide — full power: many typed questions in one round trip
Environment variables
Only the key is required — the endpoint is chosen from it automatically:
TYPESAFE_API_KEY Your Jev key (aliases: JEV_API_KEY, JEV_KEY). Required.
JEV_BASE_URL auto — chosen from your key:
jv_live_… key → https://jevtypesafeai.com/api/v1/decide
other key → https://api.typesafe.ai/v1/systemone
Set explicitly to route through your own gateway.
JEV_MODEL jev-latest (pin e.g. jev-1.13.0 in production)Gotcha: a hosted jv_live_ key only works at the gateway
A hosted jv_live_… key is valid ONLY at the jevtypesafeai.com gateway, not at api.typesafe.ai. Current jev-mcp detects a jv_live_ key and routes it there automatically, so you don't touch JEV_BASE_URL. If you're on an older version and a hosted key returns 401 authentication_error, upgrade — or set JEV_BASE_URL=https://jevtypesafeai.com/api/v1/decide explicitly. An official TypeSafe key (from console.typesafe.ai) uses api.typesafe.ai by default.
Route through your own gateway
Set JEV_BASE_URL explicitly to send calls through your own gateway instead — Vercel AI Gateway, OpenRouter, or Cloudflare. Everything else stays the same.
FAQ
Why does my Jev MCP key return 401 authentication_error?
Almost always the endpoint. A hosted jv_live_ key is valid only at the jevtypesafeai.com gateway, not at api.typesafe.ai. Upgrade jev-mcp (it auto-detects jv_live_ keys) or set JEV_BASE_URL=https://jevtypesafeai.com/api/v1/decide explicitly.
What key does jev-mcp need?
TYPESAFE_API_KEY (aliases JEV_API_KEY, JEV_KEY). Use a hosted jv_live_ key from jevtypesafeai.com/pricing, or an official key from console.typesafe.ai.
Do I need to set JEV_BASE_URL?
No, in normal use. jev-mcp picks the endpoint from your key — jv_live_ keys go to the hosted gateway, official keys to api.typesafe.ai. Set it only to force your own gateway.
Which MCP clients does jev-mcp work with?
Any MCP client — Claude Code, Cursor, Codex, OpenCode and others. Only the config file location differs; the command and env are the same.
How do I pin a model version?
Set JEV_MODEL, e.g. JEV_MODEL=jev-1.13.0, so production stays on a fixed version instead of jev-latest.
See also: Cursor integration · Claude Code integration · jev-mcp & open source
Try Jev before you wire it in
Run a real, typed decision in the browser — free, no signup — then drop your key into the example above.