SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsjevx-mcp

jevx-mcp

Run zero-shot Jev AI typed decisions with criteria defined in the request and get calibrated probabilities for every option.

Open website

About this tool

Purpose

jevx-mcp is an MCP (Model Context Protocol) server used to connect the Jev AI decision-making capabilities provided by Jevx (jevx.org) to any MCP-compatible client, such as Claude, Cursor, Cline, Codex. Its goal is to enable clients to directly initiate "typed decision" calls in a conversation or workflow, without having to handle the underlying HTTP requests themselves.

Main capabilities

  • Run Jev AI's typed decisions: the user provides a state and a question used for branch judgment, and it returns a choice, a score, or a yes/no result.
  • Each option comes with a calibrated probability, making it easy to compare among multiple candidates or make branch judgments.
  • Jev AI is a TypeSafe zero-shot decision model: input text or JSON, output typed numerical values; when the label set changes, no dataset, fine-tuning, or retraining is needed.
  • Provides tool capabilities such as decide and validate_questions, used for actual decision-making and question validation.

How to use

  • Create an API key at jevx.org/settings/apikeys. The material clearly states: there is no anonymous decision endpoint, so a key must be used, just as when running the playground without logging in.
  • Point the client to npx -y jevx-mcp(stdio transport), and set it in its environment JEVX_API_KEY.
  • When calling, pass in the state and question to obtain the choice, score, or yes/no result as well as the calibrated probabilities of each option.

Applicable scenarios

  • When you need to directly use zero-shot classification or branch decisions in an MCP client (such as Claude Desktop / Claude Code, Cursor, Cline, Codex).
  • Scenarios where the label set changes and you do not want to prepare a dataset again or retrain the model for each change.
  • Scenarios where you want to compare multiple options in the form of probabilities and assist in judging the direction of a branch.

Limitations and notes

  • You must hold and configure JEVX_API_KEY; there is no anonymous decision endpoint.
  • Regarding rejection cases: the API still returns HTTP 200 on rejection (invalid key, malformed request, etc.), but with a non-zero code; decide and validate_questions present this as an ordinary tool error rather than silently succeeding.
  • Jevx is an independent playground and API for Jev AI, and is not affiliated with TypeSafe AI.
  • The material does not provide other capabilities, fee standards, prices, or specific usage steps; this information is not currently available.

Related tools

Browse category