What it is
An MCP server that lets Claude Code, Codex and other MCP-capable AI agents drive a real Chrome browser. The agent plans; jev-mcp turns the page into a clean model and JEV makes the many small browser decisions.
How it uses Jev
JEV answers typed questions such as which element is a departure-city input or whether a click worked, on a filtered page model (DOM, accessibility tree, layout, regions, element refs). Each answer carries a calibrated confidence; the tool acts when sure and asks the agent when unsure.
Primitives:choicescorenoul
Technique worth stealing
Route small browser decisions to a fast typed System One model with calibrated confidence, escalating to the planner only when unsure.
Try it
git clone, npm install, npm run build:web, node bin/jev.mjs install, set API key, jev doctor.
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 patternyes0.94
- Handles uncertaintyunclear0.54
- Measuredno0.32
- Runnableyes0.97
- Worth recommendingunclear0.43
- Model replicano0.07
- Problem scopescore on a 0–2 scale1.15
- About Jevyes0.98
Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 23 Sept 2026.