What is reflex-cu and what does it control?

An open-source computer-use MCP server that lets Claude Code, Codex and other agents operate macOS and Windows desktops, including a remote Windows machine over SSH, by reading the accessibility or UI Automation tree.

How does reflex-cu use Jev?

High-frequency reflex judgments, such as which item to click, whether an action succeeded and whether something finished loading, are delegated to Jev, TypeSafe's System One model, instead of being reasoned about by the coding agent.

What is the look feature in reflex-cu?

Since version 0.4, find, check and wait accept look: true and Jev looks at screenshots directly using Jev 1.14 image input and boxes, to locate pure icons, characters and objects in game scenes; it remains experimental.

Which keys and platforms does reflex-cu accept?

A TypeSafe, OpenRouter or Command Code key works for the standard judge; the look feature currently requires an official TypeSafe key. Modes cover a local Mac, local Windows, or remote Windows over SSH.

Do screenshots leak into my agent context?

Per the README, screenshots taken for the look feature are sent only to Jev and do not enter the coding agent's context. The digest excerpt does not document logging or retention beyond that statement.

How do I install reflex-cu?

The digest does not include install commands: the README section shown truncates at the tools table. The repo is Python, MIT-licensed, with 2 stars, created 2026-10-09 and pushed 2026-10-15.

How does reflex-cu differ from screenshot-based computer use tools?

Instead of screenshotting at every step, it reads the accessibility / UI Automation tree, and only the experimental look feature asks Jev to view pixels for icons or game objects, keeping screenshots out of the agent's context.

Jev directory

CommunityJev Agents & MCP

Reflex Cu

An open-source computer-use MCP server for Claude Code, Codex and other agents that reads the accessibility tree instead of screenshotting every step, using Jev as a fast judge and, experimentally, as image input to spot icons and game objects.

See highlights Open github.com

Stars: 2
Language: Python
License: MIT
Reading time: 2 min
Published by: Community
Author: 8itlab
Last verified: Oct 2026

8itlab/reflex-cu

User repository on GitHub

View on GitHub

Computer use MCP server for Claude Code, Codex and other AI agents. Reads the UI tree instead of taking a screenshot every step, with Jev (TypeSafe System One) as a fast judge that can also look at the screen to find icons and game objects. Local macOS, local Windows, or remote Windows over SSH.

Forks: 0
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

Tagscommunity github mcp computer-use macos

Works forJev for Realtime control Jev for Verification Jev for Decision gating

Used inJev for Software & DevTools

Languages: Python
Platforms: MCP macOS
Capabilities: Open sourceRuns locallyRequires API key

Highlights

A computer-use MCP server for Claude Code, Codex and other agents, driving macOS and Windows desktops plus remote Windows over SSH.
It reads the accessibility / UI Automation tree instead of taking a screenshot at every step.
Jev (TypeSafe System One) acts as a fast judge for which-to-click, success and loading checks; the README cites about 0.3 s per call.
Since 0.4, look: true on find, check and wait lets Jev 1.14 see the screen to find icons and game objects.
Screenshots go only to Jev and stay out of the agent's context; the look feature is experimental and needs an official TypeSafe key.

Watch out

README (Chinese original with an English translation) and GitHub metadata reviewed 2026-10-17; the digest truncates the tools table, so the full tool list, installation steps and pricing are not evidenced here.

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: Voice-controlled computer use on a Mac

X: Computer use without screenshots

X: Jev in front of the HeyGen MCP

X: Playwright CLI + Jev vs Playwright 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