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

jevwire

Jev decision layer for agents: seven MCP tools, an embeddable library and escalate-only Claude Code hooks at harness boundaries.

MCP serverbetaMIT
View on GitHub ↗

Overview

jevwire wires Jev into an agent harness in three ways. An MCP server provides seven tools: jev_rank, jev_pick, jev_verify, jev_evaluate, jev_gate_action, jev_next_step and jev_list_models. An embeddable library, JevDecisionModel with a pure function per tool, lets mandatory checks live in your harness rather than in a tool the agent might decline to call. A Claude Code plugin places judgments at harness boundaries: before a tool call, after a fetched result and before a turn ends.

Since version 0.3.0 the hooks never prompt the user; their output is addressed to Claude as a note about a call or a single deny Claude can respond to, and judgments do not grant native permissions. Without a key, deterministic pattern checks still run. The README is explicit that Jev is not for generation, arithmetic, counting, date comparison or multi-hop reasoning, and that numeric comparisons should be extracted as choices and compared in code. Figures in the README come from a live run of 0.4.0 on 2026-09-18. It also works with Claude Desktop and Codex as an MCP server.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich option should be picked? (jev_pick)—
Yes / noShould this action be allowed? (jev_gate_action)—
Yes / noDoes the evidence verify this claim? (jev_verify)—
ChoiceWhat is the next step? (jev_next_step)—

How it works

  1. Install as a Claude Code plugin, MCP server or library
  2. Hooks or tools build a state and typed question
  3. Jev answers; pattern checks run regardless
  4. Hook output is a note or a deny addressed to the agent
  5. Agent continues or adjusts

Strengths

  • Mandatory checks can live in the harness
  • No user prompts from hooks
  • Dependency-free bundled files
  • Clear statement of what Jev should not be used for

Caveats

  • Judgment hooks inactive without a key
  • Hooks do not grant native permissions
  • Live figures from a single dated run

Getting started

/plugin marketplace add Brainwires/jevwire then /plugin install jev@brainwires-jevwire, or claude mcp add jev -e TYPESAFE_API_KEY=... -- npx -y jevwire

At a glance

Type
MCP server
Status
beta
Language
TypeScript
Runtime
MCP server, Node library, Claude Code plugin
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Provides Jev MCP tools, an embeddable library and Claude Code hooks for Agents.

    logicrw/awesome-jev-projects

  • MCP tools, an embeddable decision library, and advisory or restrictive Claude Code hooks; judgments do not grant native permissions.

    cobanov/awesome-jev

  • Jev decision layer for agents: MCP server, embeddable DecisionModel library, and an escalate-only Claude Code plugin (TypeSafe AI's Jev).

    hellogumbo/awesome-jev

  • An MCP server, an embeddable decision library, and an escalate-only Claude Code plugin in one repository.

    MrJev/awesome-jev

Frequently asked questions

Will the hooks prompt me?

No; since 0.3.0 they address notes or denies to Claude, not the user.

What should I not use it for?

Generation, arithmetic, counting, date comparison or multi-hop reasoning, per the README.

Does it work without a key?

Deterministic pattern checks still run; Jev judgments are inactive.

Related projects

  • jev-mcp

    MCP server giving any agent Jev's classify, score, yes/no check and allow/confirm/block gating of risky tool calls.

    MCP serverbetaJavaScript
  • master-jev-hook

    Local gateway, MCP server and Claude Code hooks that let coding agents delegate explicit choices, classification and scoring to Jev.

    MCP serverbetaTypeScript
  • jev-fabric

    Native Bend executable that owns agent-launched processes and logs, and asks Jev a typed question only when told to.

    ServicebetaBend
  • 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
  • jev-layer

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

    ProjectbetaTypeScript
  • jev-mcp

    Rust MCP server for Codex that hands classification, relevance, candidate selection and evaluation to Jev with accept/verify policies.

    MCP serverbetaRust