← Back to Lab
Jev Judge MCP
An MCP server that gives coding agents a judge tool for typed questions over supplied evidence, with calibrated probabilities from TypeSafe's Jev.
Technologies
JavaScriptNode.jsMCPTypeSafe Jev
Typed judgments for agents: Exposes a single judge tool through MCP, letting agents ask narrow choice, yes/no, and score questions over evidence they supply.
Probabilities and boundaries: Returns probabilities instead of generated prose. The server sends only the supplied state and questions to TypeSafe’s API, with explicit errors and cancellation support.
Portable setup: Runs as a stdio MCP server for Claude Code, Codex, OpenCode, and other MCP clients. A companion skill helps agents design useful questions.