What is mpv-agent?

mpv-agent adds an AI assistant to mpv, the open source media player, so you can control playback by typing plain language instead of memorizing keyboard shortcuts or command syntax. You open a panel inside the player and type requests like "set bottom subs to english and top subs to spanish" or "download hungarian subs" or "switch to my speakers and use the 5.1 track," and the tool works out what mpv commands to run and checks them before they touch the player. Each mpv window keeps its own separate AI conversation, so if you have several videos open at once each one remembers its own context. Behind the scenes you choose a backend: Codex, Claude Code, and OpenCode keep a running conversation with the AI provider and can use its tools within limits you set, while an OpenAI-compatible option, including OpenRouter or a locally run vLLM server, treats every request separately and returns a plan of actions that gets checked before running, with no memory kept between requests. OpenCode can also run fully local models through Ollama if you prefer not to send anything to an outside provider. mpv-agent sends your requests and basic playback details to whichever provider you pick, but not the video or audio content itself. Two settings, agent_access and mpv_access, control how much the AI is allowed to do: whether it can touch your shell, filesystem, or network, and whether changes it wants to make to mpv need your approval first. The project recommends starting in a restricted, guarded mode. You can also write a personal instructions file so the assistant remembers preferences, such as which subtitle languages you like. To use it you need mpv 0.41 or newer, Node.js 20 or newer, and either one of the supported AI command line tools or access to an OpenAI-compatible API. It installs with a single script on Windows, Linux, or macOS, and is released under the MIT license.