ClaudeMap

Jev MCP

A local MCP server that turns TypeSafe's Jev into eight typed judgment tools — verify, screen, find, rerank, classify, decide, compare, extract.

MCP Serversmcpjudgmentjevverificationclassification
Visit projectjkudish/jev-mcp

jev-mcp gives any MCP client a small judgment layer: jev_verify checks claims against evidence, jev_screen judges content before it enters context, jev_find picks the best candidate by meaning, jev_rerank scores and sorts candidates, jev_classify batch-assigns classes, jev_decide settles bounded alternatives, jev_compare judges how two passages relate, and jev_extract pulls field values with regex plus judgment. Each answer comes back typed with probabilities and a confidence score in roughly 150-500 ms, for a fraction of a cent — the cheap mechanical checks agents normally skip because a frontier model is too slow to run on every page or candidate. The README documents real catches: a contradicted claim flagged at confidence 1.0, and a prompt-injection note on a pricing page blocked at probability 0.99 while the page itself still read as genuine. npm package @jkudish/jev-mcp; registers with Claude Code, Codex, OpenCode and Amp.

What it does

  • Eight typed judgment tools: jev_verify, jev_screen, jev_find, jev_rerank, jev_classify, jev_decide, jev_compare, jev_extract — answers carry probabilities and confidence
  • Fast and cheap per call: roughly 150-500 ms and a fraction of a cent, cheap enough to run on every claim, page, or candidate list
  • One npx command registers anywhere MCP does: claude mcp add, Codex TOML, OpenCode JSON, Amp
  • README documents real catches: prompt injection blocked at probability 0.99 while the page still read as genuine

When to use it

Agents that need many small, structured judgments — claim checking, screening content before it enters context, ranking candidates — without paying frontier-model latency and cost on each one.

Getting started

claude mcp add jev -- npx -y @jkudish/jev-mcp

# requires Node.js 20+ and a TypeSafe API key (console.typesafe.ai)
# other clients: Codex config.toml, OpenCode opencode.json, Amp mcp add — same npx command

Things to know

Early software — the README says to expect rough edges. Requires a TypeSafe API key, and the content you ask it to judge goes to TypeSafe's API. Probabilities are model outputs, not guarantees — keep a human in the loop for consequential decisions. MIT; latest npm release 0.4.0 at review time.

Repository

jkudish/jev-mcp

Charted

Related on the map

  • Blitz Strike

    MCP Server

    A penetration-testing methodology delivered as one MCP server: BLITZ maps the attack surface, EAGLE-EYE traces source-to-sink, STRIKE live-verifies every finding before it is reported.

    • mcp-server
    • security
    • pentest
    • methodology
    View

  • Ripwire

    MCP Server

    Red Hat's "ripgrep for AI context" — a C++23 zero-dependency CLI that also speaks MCP.

    • mcp
    • code-context
    • cli
    • red-hat
    View

  • Deja Vu

    MCP Server

    One shared session memory for 25+ coding agents — Claude Code, Codex, and Cursor finally remember the same project.

    • mcp
    • memory
    • cross-agent
    • context
    View