Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-browse (0x7067)

Agent-plugin browser tool where Jev picks one operation and one row from a numbered page table; a small LLM only writes typed text.

Projectbeta
View on GitHub ↗

Overview

jev-browse by 0x7067 takes a goal and a start URL and drives Chrome until the goal is done or it gets stuck. On each step Jev reads a numbered table of what is on the page and picks one operation and one row; it never writes selectors, coordinates or code, so every target it names exists. A small LLM only runs for TYPE_TEXT, to write the text. A demo searches Google Flights for one-way flights from Zürich to London in about 14.6 s and 11 actions, including model calls and page loads.

The repository is an Agent Plugins package with plugin, MCP and skills definitions and a committed esbuild bundle, so nothing is built on install. It installs as a Pi tool, a Claude Code or Codex plugin, an OpenCode plugin, an MCP stdio server for any client, or a CLI. It uses TypeSafe or OpenRouter as the Jev provider and needs Node 22+ and Chrome.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich operation should be performed next?CLICK · TYPE_TEXT · ...
ChoiceWhich row of the page table is the target?—

How it works

  1. Observe the page as a numbered element table
  2. Jev picks an operation and a row
  3. For TYPE_TEXT, a small LLM writes the text
  4. Perform the action in Chrome and repeat
  5. Stop when done or stuck

Strengths

  • Targets always exist on the page
  • Installs into many agents without a build step
  • Minimal LLM use

Caveats

  • Requires Chrome and Node 22+
  • No licence stated in the excerpt

Getting started

For Claude Code: `claude plugin marketplace add 0x7067/jev-browse` then `claude plugin install jev-browse@jev-browse`; set TYPESAFE_API_KEY or an OpenRouter key.

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
agent plugin, MCP server, CLI
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

One operation and one target row from the page table at each step.

When is an LLM used?

Only to write text for TYPE_TEXT actions.

Which agents can use it?

Pi, Claude Code, Codex, OpenCode and any MCP client.

Related projects

  • jev-browser-control

    Chrome extension plus MCP server so Claude or Codex can drive your real Chrome, with Jev picking each click in about half a second.

    MCP serverbetaTypeScript
  • jev-max

    Jev browser agent that pierces shadow DOM and iframes, adds canvas fallback and full actions, with one Jev request per cycle.

    MCP serverbetaPython
  • jev-mcp (legostin)

    MCP server that lets Claude Code or Codex drive real Chrome with Jev picking elements and checking steps, plus safety gates and traces.

    MCP serverbetaTypeScript
  • jev-ra

    Fast browser-use layer for coding agents: Jev picks the operation and target element each step, via MCP server or CLI.

    MCP serverbetaPython
  • jev-ultrafast-mcp

    MCP server that takes a whole browser task (URL, goal, checks) in one call and drives Chrome server-side with a decision model.

    MCP serverbetaPython
  • jev-browser-skill

    Readable reference skill for Claude Code and Codex where Jev picks from a numbered table of page controls, with an explainer site.

    Agent skillprototypeJavaScript