jev-mcp (jkudish)
Node MCP server with a dozen fail-closed Jev judgment tools: verify, screen, find, rerank, classify, extract, review and gate.
Overview
Jev MCP by jkudish gives an agent a set of judgment tools backed by TypeSafe's Jev. They cover checking claims against evidence, screening fetched content for injected instructions before it enters context, calibrated yes/no on a proposition, picking or reranking candidates by meaning without embeddings, batch classification, settling bounded alternatives with an ask-the-user escape, comparing passages, extracting verbatim field values with regex plus judgment, auditing extracted values, reviewing a diff, and gating completion claims.
The README pitches these as the cheap mechanical checks agents usually skip because a frontier model is too slow to run on every page or claim, citing roughly 150-500 ms per judgment. Handling is fail-closed, and the npm package ships an agent skill so coding agents get a judgment policy out of the box. The author describes it as early software.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | jev_verify: is this claim supported by the cited evidence?(inferred) | yes · no |
| Yes / no | jev_screen: does this content contain injected instructions?(inferred) | yes · no |
| Choice | jev_find: which candidate best answers the question?(inferred) | — |
| Score | jev_review: how correct and complete is this diff (correctness, spec match, test gap, blast radius)?(inferred) | — |
How it works
- Install @jkudish/jev-mcp and set a Jev provider key
- The agent calls a tool such as verify, screen, find or gate
- Jev returns typed probabilities and confidence
- Fail-closed handling blocks or escalates when answers are uncertain or invalid
Strengths
- Broad toolset covering verification, screening, retrieval and review
- No embeddings or index needed for semantic finding
- Agent skill bundled for out-of-the-box policy
- Extraction returns verbatim strings rather than generated text
Caveats
- The author describes it as early software with rough edges
- Requires Node 22+ and a provider key
- Latency and cost figures are the README's own
Getting started
npx -y @jkudish/jev-mcp (register it as an MCP server with TYPESAFE_API_KEY set).At a glance
- Type
- MCP server
- Status
- beta
- Language
- TypeScript
- Runtime
- Node MCP server
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
MCP ecosystem: MCP server exposing eleven Jev judgment tools (verify, screen, noul, find, rerank, classify, decide, compare, extract, review, gate) behind fail-closed handling, with an agent skill shipped in the npm package so coding agents get judgment policy out of the box.
yibie/awesome-jevNode MCP wrapping three cookbook patterns: `jev_verify` (citation check), `jev_screen` (prompt-injection / guardrails), `jev_find` (semantic ranking without embeddings). `npx -y github:jkudish/jev-mcp`.
AnotiaWang/awesome-decision-modelsEight MCP tools for evidence checks, content screening, search, reranking, classification, comparison and extraction.
logicrw/awesome-jev-projectsFast, cheap, typed judgments from TypeSafe.
wh000wh000/awesome-jev-liveProof of concept MCP for Typesafe's new Jev AI model.
hellogumbo/awesome-jevProof-of-concept MCP server with ready-made tools for fact checking, prompt-injection detection, and semantic ranking.
MrJev/awesome-jev
Frequently asked questions
What kinds of judgments does it provide?
Claim verification, content screening, yes/no propositions, finding and reranking candidates, classification, comparison, extraction, diff review and completion gating.
Do I need embeddings for search?
No. jev_find and jev_rerank judge candidates by meaning directly.
Which provider does it use?
TypeSafe direct by default; the README lists alternatives under Configuration.
Related projects
MCP server giving coding agents typed judgment tools (verify, screen, rerank, gate, score) backed by Jev, with auto/review/escalate policy.
MCP serverbetaPythonSmall stdio MCP server with ask, rank and verify tools backed by Jev, a local event log and installers for several agent harnesses.
MCP serverbetaTypeScriptMCP server giving agents typed Jev decisions (ask, choice, score, noul, batch) with calibrated probabilities via the OpenJEV API.
MCP serverbetaTypeScriptCisco's security scanner for agent skills, with a System One analyzer as an advisory tier that cannot emit findings or change severity.
Agent skillproductionPythonVideo summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.
ProjectbetaPythonLocal MCP bridge letting Codex hand drafting to Gemini via a Google AI Pro quota, with Jev checking drafts against explicit criteria.
MCP serverbeta