Jev MCP
Ten MCP tools on TypeSafe Jev: verify, screen, find, rerank, classify, decide, compare, extract, review, and gate in about 150 to 500 ms.
Overview
jev-mcp (github.com/jkudish/jev-mcp, npm @jkudish/jev-mcp) exposes ten MCP tools backed by TypeSafe Jev: jev_verify, jev_screen, jev_find, jev_rerank, jev_classify, jev_decide, jev_compare, jev_extract, jev_review, and jev_gate. README positions each call at roughly 150 to 500 ms for a fraction of a cent, returning probabilities and confidence instead of prose essays. Install with npx -y @jkudish/jev-mcp and register the server in your MCP client; pass TYPESAFE_API_KEY explicitly when the host strips environment variables.
Problem: Agents skip cheap verification, rerank, and gate steps because calling a frontier model on every page or claim is too slow and too expensive.
Built for: Claude Code, Codex, OpenCode, and other MCP hosts that want ten typed Jev tools with explicit schemas instead of ad hoc evaluate HTTP.
First indexed on Jev Directory: 2026-09-23
Creator and team
- Name
- Joey Kudish
- Handle
- @jkudish
How Jev is used
- Role in the product flow
- MCP tool transport for parallel verify, screen, find, rerank, classify, decide, compare, extract, review, and gate judgments
- Primitives
- ChoiceScoreNoul
- State in
- Tool JSON payloads: claims with evidence text, candidate lists, label sets, diffs, or patch plus completion claims per tool schema in README.
- Decision out
- Typed verdicts, ranked lists, label probabilities, field extractions, or combined patch review plus completion gate scores with confidence and suggested actions.
- Agent registers npx -y @jkudish/jev-mcp as a local MCP server
- Each tool maps to one focused System One question batch
- Agent thresholds probabilities in code instead of parsing chat
- jev_gate merges patch review with completion claim verification in one call
Joey packages the boring agent checks as first-class tools so hosts can call them on every retrieval page, inbox row, or PR hunk without spinning up a separate microservice. jev_verify and jev_screen cover claim and prompt-injection hygiene; jev_find and jev_rerank replace embedding maintenance for many workflows; jev_review and jev_gate mirror staged code review patterns documented elsewhere in this directory but as drop-in MCP calls. classifier-dev-mcp remains the path when you want hosted classify_texts on classifier.dev rather than general judgment tools on your key.
Sourced performance claims
- README cites roughly 150 to 500 ms latency and fractional cent cost per judgment call.Source: github.com/jkudish/jev-mcp README
- Public GitHub repo jkudish/jev-mcp had 316 stars and 34 forks when this listing was drafted.Source: GitHub API September 2026
- Package name @jkudish/jev-mcp on npm with server command npx -y @jkudish/jev-mcp documented for multiple MCP clients.Source: github.com/jkudish/jev-mcp README install section
Features and stack
Features
- Ten documented MCP tools with JSON examples
- Node.js 20 plus and TypeSafe API key
- Install snippets for Claude Code, Codex, OpenCode, and Amp
- Agent-install prompt block in README for guided setup
- MIT licensed TypeScript server with CI badge
Stack
- TypeScript
- MCP
- Node.js 20+
- TypeSafe System One
Pricing: Open source server; TypeSafe usage billed per README latency class on your API key.
Links
FAQ
- How is jev-mcp different from classifier.dev MCP?
- classifier-dev-mcp wraps hosted classify_texts on classifier.dev. jkudish/jev-mcp ships general verify, rerank, gate, and review tools on your TypeSafe key.
- Why does my client say the API key is missing?
- README warns some MCP hosts filter environment variables. Pass TYPESAFE_API_KEY in the server env block explicitly.
- Is jev_gate the same as jev_review?
- jev_review scores a diff against rubric dimensions. jev_gate combines patch review with verification of completion claims in one tool per README.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev & TypeSafe
What Jev is, how System One fits your stack, and where this directory ends and TypeSafe docs begin.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.