Developers
Your agent sees the screen, moves the mouse, presses keys, and runs Luau. Drive the same input from HTTP or TypeScript, Python, and Rust.
Field demo
One prompt, start to finish. An agent opens Safari, reads Hacker News, and returns three AI stories as JSON. No browser framework. No scraping API.
01 / the task
02 / the plan
03 / the hands
04 / the result
Connect an agent
One command wires Rebind into Claude Code, Codex, or Cursor. Start the remote script in the app and your agent has a desktop.
One command, then restart the client.
npx add-mcp "bunx @rebind.gg/mcp-server" --name rebind -g -a claude-codeNeeds the Rebind app with the remote script running. Check the link: bunx @rebind.gg/mcp-server --selftest
GitHub(opens in a new tab)npm package(opens in a new tab)Documentation(opens in a new tab)
Runs on the machine you want controlled. Connection and permissions.(opens in a new tab) Need hardware output? Rebind Link sends it at up to 8,000 Hz.
The toolkit
One job per tool. Coordinates come back in the pixel space of the last screenshot, so the model never rescales.
perception
screenshotscreenshot_windowzoom
pointer
clickmove_mousescroll
keyboard
typekey
windows
list_windowsfocus_windowactive_windowlist_displays
scripting
run_lualua_docs
control
get_cursorwaitcalibrate
A local API in the Rebind app. TypeScript, Python, and Rust clients move the mouse, type, and read pixels.
// npm install @rebind.gg/client-ts
import { RebindRemote } from "@rebind.gg/client-ts";
const r = new RebindRemote("ws://127.0.0.1:19561");
await r.connect();
r.hidMove(30, -5);
r.hidPress("Mouse1", 20);
r.hidType("hello");
const { x, y } = await r.systemMouse();
const px = await r.screenPixel(x, y);
r.close();
Python and Rust clients match this surface. Install and API docs.