wtf is obs_mcp?

OBS-MCP lets an AI assistant like Claude Desktop, Claude Code, or Cursor control OBS Studio, the popular streaming and recording software, by talking to it in plain language. It connects to OBS through a local WebSocket connection built into OBS itself since version 28, so everything runs on your own computer and your stream setup never leaves your machine. The AI only reaches OBS through this tool, and the tool only reaches OBS through the connection you enable in its settings. The project exposes around 148 individual controls covering nearly everything OBS-websocket allows, including switching and creating scenes, adjusting audio levels and routing, applying and reordering filters, positioning sources, managing transitions, and starting or stopping a stream or recording. Rather than making the AI figure out every low-level setting on its own, it also includes two combined tools that handle common real-world tasks in one step. One builds a full noise gate, noise suppression, and compressor chain to clean up a microphone's audio using parameter values checked against OBS's own filter source code. The other pulls together streaming and recording statistics and interprets them, pointing out whether dropped frames are likely caused by your graphics card, your encoder, your network, or low disk space. Setting it up involves installing the tool with Python's package manager, turning on OBS's built in WebSocket server from the Tools menu, and adding a short configuration snippet to your AI client that points it at your OBS connection details, including a password if you set one. Once that is done, you can ask your AI assistant to switch scenes, start streaming, clean up your microphone, or check why your stream might be dropping frames, and it can act on those requests directly. This project requires OBS Studio 28 or newer and Python 3.10 or newer, and it is released under the Apache 2.0 license.