Adds a set of Jev tools to any MCP client: choosing the next step, reviewing a proposed change, checking claims against evidence, screening untrusted content and ranking options.
Each answer comes back with a suggested action, either go ahead, review first or hand it to a bigger model, so the agent knows when it can move on by itself.
Runs locally and needs a TypeSafe API key.
What you get
- Full TypeScript source code, MIT licensed
- Nine Jev tools and their question packs
- Examples, docs and tests






