jev-gate-mcp
MCP gateway that trims noisy tool output and offers task classification and review advice, splitting work between local Laya and Jev.
Overview
Jev Gate (MCP) helps AI coding assistants read less noise. It judges which parts of logs, search results and tool output are relevant to the current work and trims the rest in supported environments, shortens tool records while keeping kept content verbatim, and offers task classification and review suggestions for Claude Code and Codex. The README warns its opinions can be wrong and must not be treated as authorisation; completion checks still need real execution evidence.
With JEV_PROVIDER=laya, a local Laya model (about 540 MB, 1.2 to 1.9 GB memory when loaded) handles short fixed-type classifications such as task type, while Jev handles difficulty, risk, completion verification and fallback when Laya fails. Log trimming and compaction always use Jev. Laya is described as suitable only for limited trials, and savings are not yet measured. Local classifications are retained for review unless logging is turned off. Documentation is in Traditional Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this section of tool output relevant to the current task?(inferred) | — |
| Choice | What type of task is this?(inferred) | — |
| Score | How difficult or risky is this task?(inferred) | — |
How it works
- Tool output or a task arrives at the MCP gateway
- Laya tries simple classifications locally if enabled
- Jev handles relevance trimming, risk, difficulty and fallbacks
- Return trimmed output or advice to the assistant
Strengths
- Less irrelevant text in context
- Local model can save some cloud calls
- Kept content stays verbatim
Caveats
- Laya quality limited and under trial
- Savings not yet measured
- Local logs retain text by default
Getting started
`git clone https://github.com/royalskynet/jev-gate-mcp.git && cd jev-gate-mcp && npm install`, then follow the Claude Code or Codex connection guide.At a glance
- Type
- MCP server
- Status
- prototype
- Language
- TypeScript
- Runtime
- MCP server (Node 22/24) with optional local Laya
- Has code
- yes
Use cases and integrations
How curated lists describe it
Governed Jev decision gateway for GPT, Claude, and MCP clients.
wh000wh000/awesome-jev-live
Frequently asked questions
What does it decide?
Which tool output is relevant, plus task type, difficulty and risk.
How do Laya and Jev split work?
Laya does short fixed classifications locally; Jev handles everything else and fallbacks.
Can its advice authorise actions?
No, the README says it can be wrong and is not authorisation.
Related projects
Local Python MCP service that keeps the skill catalog out of the agent's context and lets Jev pick which SKILL.md files to load.
MCP serverbetaPythonOne-command npx installer for a set of Jev-powered agent tools: context compaction, browser agent, prompt enhancer and skill router.
ProjectbetaJavaScriptClaude Code, Codex and pi plugin that hands non-writing agent steps to Jev and escalates writing or unsure steps to the LLM.
PluginbetaJavaScriptMCP server, CLI and Claude Code skill with Choice, Score and Noul primitives and a confidence gatekeeper, hosted or offline.
MCP serverbetaTypeScriptSelf-hosted Go gateway for LLMs and decision models that prunes context, and audits and calibrates Jev or open-rlcd decisions.
ServicebetaGoRust engine that answers noul, choice and score from any LLM in one prefill via logprobs, wire-compatible with /v1/systemone and MCP.
ModelbetaRust