jev-classifier
Local gateway and MCP tools that let Jev predict which tool a coding agent should call next, from suggestions to enforced choices.
Overview
jev-classifier connects coding agents to Jev, used here to predict which tool should be called next. You start by watching its suggestions, then where supported let it choose the tool the agent calls, and every decision is logged for review. You keep using the agent's own model and login, configuring a separate Jev key during setup. It credits jev-eval-agent by vinilana as the inspiration and brings that approach to a local gateway and MCP tools.
It connects in two ways. As a proxy, requests from Codex, Claude Code, Grok Build and OpenCode pass through the gateway on the way to the model; Codex Responses Lite records suggestions without forcing tools. As an MCP tool, Cursor and Antigravity call Jev when they want a suggestion and decide whether to follow it. It runs on Windows, macOS and Linux with Node 22+, installs globally from npm with optional desktop startup entries, and has monitoring and settings documentation. It is MIT licensed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which tool should the agent call next? | — |
How it works
- Install and configure a Jev key
- Connect the agent via proxy or MCP
- Jev predicts the next tool
- Observe suggestions, then optionally enforce
- Review the decision log
Strengths
- Gradual adoption from observation to control
- Keeps the agent's own model and login
- Logs every decision
Caveats
- Enforcement support varies by agent
- Tool prediction quality unmeasured in the README excerpt
Getting started
`npm install --global jev-classifier`, then `jev-classifier --help`.At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- local gateway, MCP server
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A local Jev tool-routing gateway for coding Agents, with an MCP suggestion interface.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Which tool the coding agent should call next.
Does it replace my agent's model?
No, the agent keeps its model; Jev only suggests or selects tools.
Which agents are supported?
Codex, Claude Code, Grok Build and OpenCode via proxy; Cursor and Antigravity via MCP.
Related projects
Open-source Jev skills for Claude Code, Codex, Cursor and Grok: one client to gate, route, guard, verify, triage, review and act.
ProjectbetaTypeScriptOrchestrator that probes which coding agents are installed and logged in, then has Jev pick the best one for a task.
Agent skillbetaPythonMCP server that lets Claude Code or Codex drive real Chrome with Jev picking elements and checking steps, plus safety gates and traces.
MCP serverbetaTypeScriptFourteen opt-in Claude Code and Codex skills where Jev makes one bounded choice: which skill, model, test, file or next step.
ProjectbetaPythonToolkit where Jev condenses large MCP responses for Claude Code and maps Codex requests to reviewed, fixed repository operations.
ProjectbetaTypeScriptGives an agent one tool that uses Jev to search a nested capability tree and return only the documented commands a task needs.
ProjectprototypeTypeScript