Jev MCP server

This is a small helper that connects your AI assistant to Jev, a decision model made by TypeSafe. Instead of writing sentences, Jev takes a question with a fixed list of answers and returns numbers: which option is most likely, how confident it is, and how the chances split across all options. It is handy when you have to make the same small judgment over and over, like sorting incoming tickets or grading risk on a list of changes.

What this MCP server does

You ask your AI a question that has a clear set of possible answers. The AI turns that into a structured question and sends it to Jev through this helper. Jev looks at the options and returns a pick, a confidence value, and a probability for each option. Your AI then reads those numbers and tells you the result in the chat. It is quick, roughly half a second per call, and each call costs a tiny fraction of a cent.

Click to zoom

What you can do with it

• Pick the most likely option from a list of two to a hundred choices
• Grade something on an ordered scale, like risk or severity
• Answer a yes/no question with a degree between 0 and 1
• Compare two candidates and show which one wins and by how much
• Check whether a piece of evidence supports a claim
• Classify up to 100 items against one shared set of labels in a batch
• Store your TypeSafe key locally so you do not have to put it in a config file

Try asking your AI

• “Which of these rollout plans is safest? Show me the probability split.”
• “Grade the regression risk of rewriting the auth middleware in place, no tests yet, on a minor/moderate/severe scale.”
• “Does this CI log support the claim that all tests passed?”
• “Sort these 60 support emails into billing, bug, feature request or other.”

What it gives back to you

You get numbers back, not paragraphs. For a choice question you see the winning option, a confidence value, and the probability for every option, so near-ties are visible. For a score question you get a fractional number and the nearest level. For a yes/no or verify question you get a value between 0 and 1 and a short verdict like not supported. Your AI may add its own explanation, but that reading is the assistant's, not Jev's.

Before you start

Who it's for

People who need to make the same small judgment many times, like triage, routing, grading or gating work, and want it fast and cheap.