jev-mcp-server

中文文档 | CI

MCP server for Jev — TypeSafe's System One model: the three official question types (choice / score / noul), plus compare, verify, batch classify, and a one-command client installer.

Jev returns typed decisions with calibrated probabilities in well under a second for a fraction of a cent — the cheap mechanical judgments (triage, routing, scoring, gating) that a frontier model is too slow and too expensive to run on every candidate, claim, or list.

Quickstart

# 1. Get a key: https://console.typesafe.ai/settings/keys

# 2. Install into your client — one command, config written for you:
uvx jev-mcp-server install claude-code    # or: pi | cursor | opencode | codex

# 3. Restart the client if running, then ask:
#    "Which of these rollout plans is safest? Give me the probability split."

No uv yet? curl -LsSf https://astral.sh/uv/install.sh | sh (or brew install uv / pip install uv).

{
  "mcpServers": {
    "jev": {
      "command": "uvx",
      "args": ["jev-mcp-server"],
      "env": { "TYPESAFE_API_KEY": "apikey_xxx" }
    }
  }
}

[mcp_servers.jev]
command = "uvx"
args = ["jev-mcp-server"]

[mcp_servers.jev.env]
TYPESAFE_API_KEY = "apikey_xxx"

Tools

Why Jev, why this server

Jev outputs decisions, not strings: a typed answer plus calibrated probabilities and a confidence value — never an explanation. That is why a call costs ~$0.00001–0.0001 and lands in ~0.5–1s, where a frontier LLM takes seconds and costs 100×+ for the same judgment. (Any "reason" your assistant adds is its own interpretation of the numbers — treat it as a hypothesis.)

Measured on real usage (single calls, indicative only):

* at $42 / 1B input tokens.

This server maps the official System One API faithfully (same three question types, validated responses, retry on 429/503/529), adds batching, caching, and the installer, and stays a single small Python package with zero dependencies beyond mcp and httpx.

Configuration

With JEVMCP_CACHE=1, identical question payloads are answered from disk at zero cost (usage.cached: true); classify deduplicates repeated items automatically.

FAQ

Why does Jev never explain its choice? By design — "decisions, not strings" is the product. The probability distribution is the output; explanations cost the latency and tokens this model exists to avoid.

Do I need another LLM or local Jev install? No. Jev is a cloud API — no local model, no helper LLM. Your agent's main model already handles when to call these tools and how to read the numbers.

Is there another Jev MCP? Yes — jkudish/jev-mcp (Node/npm) offers ten workflow-shaped tools; a Go server also exists. This one is the Python/uvx side: faithful official question types, one-command install, bilingual docs, MIT.

License

MIT