What is the jev-max browser agent?

A Jev browser agent that handles the walls other Jev agents cannot: shadow DOM, same-origin iframes and canvas-only pages. One cheap Jev call per decision cycle picks the operation and the element, keeping the speed and cost profile of typed decisions.

What does the piercing snapshot do?

It walks open shadow roots and same-origin iframes into a single indexed element table with each control's role, name, value and bounding box; cross-origin frames are recorded as regions instead of being silently dropped.

What happens on canvas-only pages?

When the element table has no usable target, a vision model grounds the instruction to coordinates and the real mouse clicks instead. Targets always resolve from observed nodes, so model output never becomes selectors, coordinates or JavaScript.

How does it decide and verify actions?

Each cycle sends one Jev request carrying the operation question plus speculative target heads; only the head matching the chosen operation can execute, covering click, fill, select, check, upload, drag, scroll, wait, done and blocked. An independent Jev check confirms the outcome before stopping.

How do I try jev-max?

Install with pip install jev-max and playwright install chromium, add a TYPESAFE_API_KEY or OPENROUTER_API_KEY plus TEXT_MODEL_API_KEY to .env, then run a goal with jev-max run; jev-max snapshot works with no API key at all.

Can other agents drive it?

Yes: jev-max mcp starts an MCP server with jev_start, jev_tick and jev_close tools for any agent host such as Claude Code, Cursor, Windsurf or Codex, and setting JEV_HEADLESS=0 lets you watch it work.

What does it cost, and what are the limits?

Jev is $0.042 per million input tokens with free output on TypeSafe and OpenRouter, so a typical 10-step task costs a fraction of a cent in decision tokens; text for FILL and SELECT comes from your own small model. The README publishes no benchmark numbers for its task suite.

Jev directory

CommunityJev Agents & MCP

Jev Max

A Jev browser agent that breaks the three walls of shadow DOM, iframes and canvas: a piercing snapshot, vision-grounded fallback clicks, uploads, drag-and-drop and multi-tab, all driven by one Jev request per cycle. It ships as a Python library and an MCP server.

See highlights Open github.com

Stars: 3
Language: Python
License: MIT
Reading time: 2 min
Published by: Community
Author: saloni-garg
Last verified: Oct 2026

saloni-garg/jev-max

User repository on GitHub

View on GitHub

The Jev browser agent that handles the real web: shadow DOM and iframe piercing, canvas fallback, uploads, multi-tab, MCP for any agent host.

Forks: 1
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github browser mcp playwright

Works forJev for Realtime control Jev for Verification

Used inJev for Software & DevTools

Languages: Python
Platforms: MCP Playwright CLI
Capabilities: Open sourceRequires API key

Highlights

Piercing snapshot walks open shadow roots and same-origin iframes into one indexed element table; cross-origin frames become regions with a canvas fallback.
One Jev request per decision cycle carries the operation question and every speculative target head together, and only the chosen head can execute.
Never-trust-DONE: an independent Jev check confirms the outcome before the agent stops, and Jev never generates free text.
Occlusion-checked clicks run shadow- and frame-aware hit testing first; covered controls are reported, never blindly clicked.
Use it as a standalone Python library or an MCP server with jev_start, jev_tick and jev_close tools for Claude Code, Cursor, Windsurf and Codex.

Quickstart

bash

pip install jev-max
playwright install chromium
jev-max run --url https://en.wikipedia.org/wiki/Main_Page --goal "Find and open the Wikipedia article about Godel's incompleteness theorems." --record-dir traces/wiki

Watch out

README reviewed 2026-10-17; the head-to-head table against other Jev browser agents is the author's own summary, the README publishes no benchmark numbers from its multi-task suite, and the project is not affiliated with TypeSafe, OpenRouter or browser-use.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

X: Playwright CLI + Jev vs Playwright MCP

X: Text-to-screenshot (Malay Vasa)

X: Google's MediaPipe Decision Maker, tested in the browser

X: Jev in front of the HeyGen MCP

From the Jev guides

Original write-ups on what this entry is for.

MCP tool routing: Jev in front of the tool call

Oct 16, 2026

Tool menus can outgrow the context that reads them. The catalog's Jev-based routers and gates narrow the list: benchmarked tool search, per-tool scoring, fail-closed approvals, and WebMCP in the browser.

Jev Agents & MCP#mcp#tool-routing#agents

Read article

Approval policies for coding agents: hooks, guards, and gateways

Oct 16, 2026

How the Jev guard family turns coding-agent permissions into auto-approve, deny, or ask: hooks that enforce, guards that fail safe, and gateways that gate generation.

Jev Agents & MCP#coding-agents#guardrails#hooks