Jev
Provides a local, dependency-free MCP server to integrate TypeSafe's Jev model with AI agents for structured decision-making.
jev-mcp offers a lightweight, dependency-free MCP server written in Python, specifically designed to bridge TypeSafe's Jev model with various AI agents such as Claude Code, Codex, and Hermes. Instead of generating text, Jev empowers agents to make precise, structured decisions—like classifying, routing, or measuring—by accepting textual or JSON states and returning typed values (choice, score, noul) complete with probabilities and confidence. This direct, typed output streamlines agent workflows by eliminating the need for complex LLM prose parsing, making decision consumption by code straightforward and efficient.
主な機能
01Delivers typed decision outputs (choice, score, noul) with probabilities and confidence
02Integrates TypeSafe's Jev model as agent tools (`ask_jev`, `list_jev_models`)
03Optimized for parallel evaluation of multiple questions per single API call
040 GitHub stars
05Robust API key handling to prevent environment variable filtering issues
06Dependency-free Python MCP server for easy integration
ユースケース
01Classifying and routing support tickets or messages based on urgency and type
02Measuring sentiment, risk, or alignment on a defined scale for text inputs
03Automating data-driven decisions within AI agent workflows without parsing LLM prose