What it is
An MCP stdio server exposing guarded_search, guarded_fetch, and research_report for agents that research the web. It annotates or gates search results and fetched pages, and streams every decision to a local dashboard.
How it uses Jev
One Jev call per search result judges relevance, credibility, and worth-fetching; one call per fetched page judges answered, content class, and steering risk. Scores drive annotate or gate mode, are written to a JSONL ledger, and stream to the dashboard over SSE.
Primitives:choicescorenoul
Technique worth stealing
Judge every research step, not a sample, using atomic observable-signal criteria and a fail-open unscored badge instead of a misleading 0%.
Try it
Run SearXNG with JSON enabled, clone jev-client and jev-scout, npm install, then register the stdio server with Claude Code.
Evidence
Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.
- Jev-centricyes0.94
- Shows a System One patternyes0.91
- Handles uncertaintyno0.07
- Measuredno0.20
- Runnableyes0.97
- Worth recommendingunclear0.55
- Model replicano0.07
- Problem scopescore on a 0–2 scale1.31
- About Jevyes0.99
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.