jev-reach

chrome-devtools-mcp plus one 'reach' tool: Jev walks the browser to the right spot so the agent makes a single devtools call there.

MCP serverbeta

View on GitHub ↗

Overview

jev-reach is chrome-devtools-mcp with one extra tool. reach walks the browser to the spot you want to inspect, using TypeSafe's Jev to pick each click, so your agent makes one devtools call there instead of one model turn per click. Verifying a frontend bug often starts with a walk (dismiss the cookie banner, log in, open settings, expand a row), and with a plain browser MCP each step is a full turn that reads a page snapshot into context.

With reach, the agent states the goal once and gets back a status and step trace, then calls list_console_messages, list_network_requests, take_snapshot or take_screenshot on the same tab; the walk is recorded by the stock collectors because it happened on the same page objects. Jev picks one element and one action from a typed list in about 300 ms. Everything in chrome-devtools-mcp keeps working unchanged. It installs as a Claude Code plugin or replaces the chrome-devtools entry in any MCP client, and supports isolated, persistent and attach modes.

What the decision model decides

Type	Question	Answers
Choice	Which element and action move the browser toward the goal?	—

How it works

The agent calls reach with a goal
Jev picks one element and action per step
The walk runs in the same Chrome the devtools tools observe
The agent makes one devtools call at the destination

Strengths

Saves model turns and context on navigation
Keeps all stock chrome-devtools-mcp tools
Console and network logs capture the walk

Caveats

Must replace, not run alongside, an existing chrome-devtools-mcp
Requires a TypeSafe key and Chrome
Node 20.19+ or 22.12+

Getting started

/plugin marketplace add rashedInt32/jev-reach then /plugin install jev-reach@jev-reach, or claude mcp add chrome-devtools -e TYPESAFE_API_KEY=... -- npx -y jev-reach --isolated

At a glance

Type: MCP server
Status: beta
Language: TypeScript
Runtime: MCP server / Claude Code plugin
Has code: yes

Use cases and integrations

Agent Decisions340

Claude Code373

How curated lists describe it

Chrome-devtools-mcp plus one tool: Jev walks the browser to the right spot, then your agent makes a single devtools call there.
hellogumbo/awesome-jev

Frequently asked questions

What does Jev decide?

Which element to act on and what action to take at each step of the walk.

Does it replace chrome-devtools-mcp?

It includes all its tools plus reach; remove the old entry to avoid two Chromes.

Will console and network logs include the walk?

Yes. The walk happens on the same page objects the collectors observe.

Related projects

public-browser
MCP server and libraries that drive Chrome over CDP and hand a decision model like Jev a bounded menu of page actions.
ProjectproductionTypeScript
ego-jev
Agent skill porting jev-ultrafast's loop into ego lite: one ~0.4 s Jev call per DOM step picks operation and target.
Agent skillbetaTypeScript
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.
ProjectbetaTypeScript
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