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

jev-mcp (burnigtm)

Local MCP server putting Jev in the coding loop: route next steps, pick prepared tool calls, review patches and verify claims.

MCP serverbeta
View on GitHub ↗

Overview

This jev-mcp is a local stdio MCP server that gives Cursor, Codex and other MCP clients typed Jev judgments while the host agent still edits files and runs commands. Its nine tools cover routing the next step, deciding whether a generative partner model is needed, selecting an exact host-prepared tool call without generating arguments, reviewing a patch, verifying claims against evidence, screening untrusted content, ranking candidates and asking custom questions.

Every result includes an action of auto, review or escalate. Incomplete context never permits auto, and an executable call is returned only for confident, suitable selections with validated host facts. The goal is fewer partner-model turns: jev_step answers routing and call selection in one request. The README states that this reduces unnecessary generative handoffs by policy but that live quality and cost savings have not been measured. Question packs are exposed as MCP resources.

What the decision model decides

TypeQuestionAnswers
Choicejev_coding_loop: what kind of handoff does the next step need?tool use · context gathering · review · user input · stop · partner model
Choicejev_tool_route: which host-prepared call should run next?—
Yes / nojev_verify: are these claims supported by the supplied evidence?(inferred)yes · no
Scorejev_review: how acceptable is this proposed patch?(inferred)—

How it works

  1. Host code prepares exact tool calls from an existing plan
  2. jev_step routes the step and selects among prepared calls in one request
  3. Result carries auto, review or escalate
  4. Host executes only confident, complete selections, then routes again
  5. A partner model is invoked only when partner_model.required is true

Strengths

  • Can save a host-model turn per loop step
  • Jev never invents arguments or executes tools
  • Empty or ineligible candidate lists return locally with zero usage
  • List sources mention around 20 test files

Caveats

  • The README says live quality and cost savings have not been measured
  • Confidence measures model certainty, not factual truth
  • Requires a TypeSafe API key and a local build

Getting started

With Node 20+, run npm ci && npm run build, export TYPESAFE_API_KEY, then node dist/index.js doctor.

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
Node MCP server (stdio)
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

  • MCP ecosystem: server that puts Jev into the coding loop for Cursor, Codex, and any MCP client, with 20 test files behind it.

    yibie/awesome-jev

  • MCP server that puts TypeSafe Jev on the coding loop in Cursor, Codex, and any…

    wh000wh000/awesome-jev-live

  • MCP server that puts TypeSafe Jev on the coding loop in Cursor, Codex, and any MCP client.

    hellogumbo/awesome-jev

  • MCP server whose tools route the next step and decide whether a partner model is needed, for Cursor, Codex, and any MCP client.

    MrJev/awesome-jev

Frequently asked questions

What does this server decide?

The next step in a coding loop, which prepared tool call to run, whether a partner model is needed, and patch and claim checks.

Does it run tools itself?

No. It only returns a selection; the host executes it.

Does it save money?

It is designed to reduce generative handoffs, but the README says savings have not been measured.

Related projects

  • jev-layer

    Portable decision layer for agent harnesses: Jev picks from host-supplied capabilities while the host keeps permissions and execution.

    ProjectbetaTypeScript
  • `kaijia323/dsh-plugin-jev`：DeepSeek Harness 的 Jev 插件（与本库同环境）

    DeepSeek Harness plugin that registers a native jev_decide tool and declares its own policy for when the agent should ask Jev.

    PluginbetaJavaScript
  • Jev DSH Decision Engine

    Decision plugin for agent harnesses where Jev recommends tools, skills or owners and scores output, for DeepSeek Harness and iPolloWork.

    Agent skillbetaTypeScript
  • jev-flash-router

    MCP server exposing Jev Noul, Choice and Score via OpenRouter so coding agents can make quick decisions without reasoning tokens.

    MCP serverbetaTypeScript
  • jev-skill-router

    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 serverbetaPython
  • 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