Skip to content

Jev Agent Kit

jevkit is a single dependency-free Node binary exposing eleven Jev-backed decisions as both shell commands and MCP tools, including guard for gating an action, plus route, triage and semantic grep over log files.

Stars
2
Language
JavaScript
License
MIT
Published by
Community
Author
walidboulanouar
Last verified
Oct 2026

About this project

Small command line and MCP tools that give agents fast, typed decisions from TypeSafe's Jev model. One binary, no dependencies, Node 18 or newer. Jev does not write text. It answers questions about text with a number: a probability for yes or no, a pick among options with a confidence, or a score between levels. That makes it a good judge inside an agent loop: cheap, fast, and easy to branch on. This kit wraps it in eleven MCP tools an agent can call, and the same tools work in a shell pipeline.

Quoted from the project's README. Read the full README.

Highlights

  • Eleven MCP tools map to CLI commands: check, choose, score, judge, route, triage, guard, grep, rank, compact and a raw jev_ask escape hatch.
  • Exit codes are split on purpose: 1 for a definite no, 2 for guard's ask, 3 for errors, so a set -e script does not read a denial as a crash.
  • The MCP batch tools take a path and read the file server-side, restricted to text and log extensions under the startup folder or JEV_ROOT.
  • As a Claude Code PreToolUse hook, guard prints ask or deny and never allow, so a model probability cannot skip Claude's own prompts.

Watch out

jev-latest is an alias TypeSafe can move to a newer version, so pin a version id if you need stable behavior, and JEV_API_URL must be https except on localhost. In hook mode every matching tool call costs one API request, and the README says keep the matcher narrow; in that mode there is no user request to compare against, so the off-task check is inactive. Exit codes need care in scripts: guard --hook always exits 0 because the decision travels in the JSON it prints, and triage, rank and compact also always exit 0 on success, so you check output rather than status. The README we reviewed is cut off mid-sentence in its behavior list, so the full key-handling guarantees and the complete retry description are not documented here; what is stated is that the key is read from TYPESAFE_API_KEY, JEV_API_KEY or a file under the user's config directory, never printed or written by the tool, and never sent over plain http to a remote host. The six use cases ship with sample data and the results shown are from one run on one machine; no accuracy figures, latency measurements or cost per call are published. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

118GitHub stars
A Claude Code skill that moves bulk judgment off Claude and onto Jev. You filter files, classify items, rank candidates or locate lines, and it returns typed verdicts, a shortlist with borderline markers, and a receipt of the tokens Claude did not read.
Jev Agents & MCP#community#jevbest
35GitHub stars
Diff-aware test selector that sends the git diff to Jev with one question per test, then prints the native filter flags and hands them to vitest, jest, node, bun, Playwright, cargo, or go test.
Jev Agents & MCP#community#jevbest#catalog-import
7GitHub stars
Claude Code plugin that scores pending tool calls and finished answers with Jev noul probabilities, denying the call with the correction attached when a rule threshold is met.
Jev Agents & MCP#community#awesomejev#catalog-import
9GitHub stars
MCP tool router for Codex that keeps large external tool inventories behind a small find_tool surface and asks Jev to select from names and descriptions at a 0.90 threshold, falling back to bounded search.
Jev Agents & MCP#community#jevbest#catalog-import
9GitHub stars
Hermes Agent plugin exposing Jev as four coding-workflow tools for parallel judgments, binary checks, option routing and rubric scoring, with a bundled question-design skill.
Jev Agents & MCP#community#jevbest#catalog-import
LangChain deepagents middleware that replaces SkillsMiddleware with per-turn judging, loading only the SKILL.md files a turn needs so large catalogs stay out of the prompt.
Jev Agents & MCP#community#jevbest#catalog-import
Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai