Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-mcp

MCP server exposing Jev-style decision models served locally by Ollama's /v1/systemone as classify, yes/no and score tools.

MCP serverbeta
View on GitHub ↗

Overview

This jev-mcp exposes Jev-style System One decision models as MCP tools so any chat model, including small local ones, can delegate fast decisions. By default it talks to Ollama's /v1/systemone endpoint (Ollama 0.35 or newer) with open decision models such as nimble (9B), tev1 (4B) and tev1:0.8b. Any server speaking the same protocol, including TypeSafe's API, works through JEV_BASE_URL.

It offers three read-only tools that each take 1-100 texts judged independently: classify with 2-26 labelled options, yes_no returning a probability, and score over 2-26 ordered levels. Results come back in input order with usage, and a failing text returns an error without failing the batch. The README suggests using it for batches, thresholdable numbers, and helping small local chat models with decisions they would get wrong.

What the decision model decides

TypeQuestionAnswers
Choiceclassify: which option (2-26 labels) fits each text?—
Yes / noyes_no: probability that the answer to the question is yesyes · no
Scorescore: which level (lowest to highest) does each text reach?—

How it works

  1. Install Ollama 0.35+ and pull a decision model such as nimble
  2. Build the server and register it in an MCP client
  3. The chat model sends up to 100 texts with a classify, yes_no or score request
  4. Results return per text in input order with usage

Strengths

  • Runs fully local with open decision models
  • Batch up to 100 texts per call
  • Per-text error isolation
  • Works with remote /v1/systemone servers too

Caveats

  • Depends on Ollama's systemone endpoint and open model quality
  • With a remote base URL, texts leave the machine
  • No licence stated

Getting started

ollama pull nimble, then npm install && npm run build, then claude mcp add jev -- node /absolute/path/to/jev-mcp/dist/index.js

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
Node MCP server
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

Does it need a TypeSafe key?

No. By default it uses local open decision models served by Ollama; a key is only needed for remote endpoints.

Which models does it use?

Open models such as nimble, tev1 and tev1:0.8b via Ollama, or any /v1/systemone server.

When is it worth using?

For batches, for probabilities you can threshold, and to help small chat models with decisions.

Related projects

  • jev-rs

    Rust engine that answers noul, choice and score from any LLM in one prefill via logprobs, wire-compatible with /v1/systemone and MCP.

    ModelbetaRust
  • decidealot

    Docker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.

    MCP serverbetaPython
  • noulxp

    NoulXP: open standard and reference engine for packaging calibrated System One decision models so one runtime can run them all.

    ProjectbetaPython
  • opendecider

    Open calibrated decision models from 400M to 80B with Jev-compatible server, MCP server and clients, benchmarked against Jev.

    ModelbetaPython
  • jev-x-kit

    MCP server, CLI and Claude Code skill with Choice, Score and Noul primitives and a confidence gatekeeper, hosted or offline.

    MCP serverbetaTypeScript
  • typesafe-mcp

    Go CLI and MCP connector exposing an evaluate tool for Jev and other System One models to Claude Code, Claude Desktop, Codex, Hermes and pi.

    CLIbetaGo