👋 Introduction
MCP server for Google TV — control paired devices (keys, text, app launch) from an AI assistant
Proof of concept — experimental and unpublished. Tool names and behaviour may change. Run it from source; don't rely on it in production.
mcp-gtv exposes Google TV control to any MCP client, so an assistant can open apps, control playback, type into search fields, and read back what's currently on screen.
🔌 Zero credentials, shared with the rest of the stack
mcp-gtv holds no tokens of its own. It reads your already-paired devices from
~/.config/gtv/config.json — the same registry gtv-cli
writes — through the @kud/gtv core. Pair once with the CLI and
the server can drive that TV with no extra setup. Every control tool returns the
TV's resulting state (power, volume, foreground app), so the model can confirm its
action landed.
🧰 What it can do
- Switch devices — list paired TVs and change the active target mid-session.
- Full remote — send any key: navigation, media, volume, power, input.
- Type text — inject arbitrary strings into the focused field via IME, no keycode mapping.
- Launch apps — open Netflix, YouTube, Prime Video, Spotify, and more by name, or pass a raw deep-link URI.
🚀 Run it (from source)
Not yet on npm — clone the repo, run it from source, and register it with your MCP client. See the repository's Development section for the current steps.