What it does
judgevet lets you send content to TypeSafe's Jev model with a question, such as whether a support ticket concerns billing, and receive a typed answer. Yes/no questions return a Noul value, which reports the model's probability of yes. Choice questions return a selected label with per-label probabilities and confidence, and score questions return a continuous score against an ordered rubric. Your application can inspect these values or check them against a local acceptance policy, such as a minimum value. The library, a CLI that returns JSON or a policy exit status, and an optional MCP server that exposes the question tools over stdio share the same contract. The official TypeSafe Python SDK offers typed questions and clients too; judgevet adds the shared library, CLI and MCP layer with local policy evaluation.
Use it when you want a service-backed judgment on text inside a Python application, a shell script or an agent toolchain, and you need the result in a structured form. The library and CLI install without the MCP runtime, so MCP is only needed when an agent host should call the tools directly.
Running it requires Python 3.12 or newer and a TypeSafe API key, supplied through the process environment or a secret provider rather than in source code. Each call sends the state, the questions and the selected model to the configured service, so only content you are authorized to disclose should be sent. The project is marked as a draft in its documentation.
Limitations and requirements
Requires Python 3.12 or newer and a TypeSafe API key. Each call sends the content, questions and selected model to the configured TypeSafe service. Only content the user is authorized to disclose should be sent. The project is unofficial and not affiliated with TypeSafe. Documentation status is draft. Live verification covers only the observed cases and the jev-1.13.0 model; other models and the 429/529 error bodies have not been observed. Diagnostic output such as protocol errors, CLI error envelopes and tracebacks can contain sensitive content and should be reviewed before sharing.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
io.github.Alberto-Codes/judgevet
Ask TypeSafe Jev yes/no, choice and score questions and receive structured model judgments.
v0.23.0As shown on registry.modelcontextprotocol.io when we last read it