About oriel-vnc
Operates any desktop with a VNC server (macOS, Linux, Windows) through screenshots, clicks and keys
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
uvx oriel-vncORIEL_VNC_HOSTSPath to the hosts file that lists each machine's VNC address and login. Default: ~/.config/oriel-vnc/hosts.jsonLocal servers need a runtime on your machine (e.g. Node.js 18+). Config file locations differ by client; see your client’s docs.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEoriel-vnc
oriel-vnc is an MCP server that lets an AI assistant operate any computer that runs a VNC server, through screenshots, clicks and keys.
It is the second MCP server in oriel, for what oriel.exe can't reach: Macs,
Linux machines, and screens with no accessibility tree, such as games, remote desktops and canvases. For Windows apps,
use oriel.exe: it reads each window as text, so the assistant doesn't need a screenshot at every step.
oriel-vnc runs on the computer where the assistant runs and connects to each machine's VNC server. Nothing is installed on the machines themselves.
Set up
-
Turn on a VNC server on each machine: Screen Sharing on macOS (System Settings → General → Sharing), a VNC server such as TigerVNC or x11vnc on Linux, TightVNC or similar on Windows. Allow it only from your own network or tailnet.
-
List the machines in
~/.config/oriel-vnc/hosts.json(or the file$ORIEL_VNC_HOSTSnames). See hosts.example.json:{ "mac": { "address": "192.168.1.20::5900", "username": "env:MAC_VNC_USER", "password": "env:MAC_VNC_PASSWORD" }, "linux-box": { "address": "linux-box.local::5901", "password": "cmd:op read op://Private/VNC linux-box/password" } }addressishost::port.username(macOS logins need one) andpasswordcan be the value itself,env:NAMEto read an environment variable, orcmd:...to take the output of a command such as a password manager's CLI. Prefer the last two, so the file holds no passwords.encodingiszrle(the default),hextileorraw. Userawif screenshots time out against a TigerVNC server.cmd_keyis the keysym thatcmdinvnc_keysends (defaultsuper).
-
Add it to your MCP client. With uv installed:
claude mcp add --scope user oriel-vnc -- uvx oriel-vncOther clients: the command is
uvxwith the argumentoriel-vnc. From a checkout,uv run --script vnc/oriel_vnc.pyworks too.
Then ask the assistant to take a screenshot of one of your machines.
Tools
| Tool | What it does |
|---|---|
vnc_hosts | Lists the configured machines, their address and last seen screen size |
vnc_screenshot | The whole screen, scaled down so its long edge is at most 1456 px |
vnc_zoom | A region at full resolution, for small text |
vnc_click | Click at x,y (left, right or middle; clicks=2 for a double-click) |
vnc_move | Move the pointer, for hover menus and tooltips |
vnc_drag | Press, move in steps, release |
vnc_scroll | Wheel clicks at x,y in any direction |
vnc_type | Type text into the focused field |
vnc_key | A key or combination: enter, cmd+space, ctrl+alt+t, win+r |
Every x,y is in the pixel space of the latest full screenshot of that machine; oriel-vnc scales it to the real screen. Actions reply with a fresh screenshot about 0.6 s later, so the assistant sees what happened.
How it behaves
- One connection per machine, kept open between calls. A failed call drops the connection and retries once.
- Quick to fail. A machine that is off or asleep fails within 5 seconds instead of waiting out the VNC client's 30-second timeout.
- Logins in memory only. oriel-vnc reads them on first use and never writes them to disk.
Safety
oriel-vnc can do anything a person at that screen could, including on the lock screen. Point it only at machines you own, keep VNC off the public internet, and don't let the assistant act on instructions it reads on screen. oriel's SECURITY.md covers how to report a problem.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.