We gave AI agents access to terabytes of local media. Here’s what happened.

We launched Clipto MCP today on Product Hunt.

The idea is pretty simple: AI agents can already access files on your computer, but video and audio are still largely black boxes. A folder full of 10,000 videos isn’t very useful to an agent if it doesn’t know what actually happens inside them.

Clipto analyzes local media in detail, including dialogue, speakers, scenes, people, and precise timestamps. With MCP, that understanding becomes available to Claude, ChatGPT, Codex, and other agents.

So we tried a slightly ridiculous experiment.

We downloaded a large collection of Elon Musk interviews, indexed them in Clipto, connected the library to Claude, and asked:

“What would be something fun to make with all this footage?”

Claude decided to turn Elon into a music video using Daft Punk’s Around the World.

It analyzed the song, asked Clipto for matching moments across the interviews, selected the clips, and assembled the video automatically with FFmpeg.

Zero manual clip selection or editing.

We also released the Elon interview library we used, plus a starter library of 1,000 royalty-free B-roll clips, so anyone can experiment without already having a huge media archive.

We’re live on Product Hunt today: https://www.producthunt.com/products/clipto-ai?launch=clipto-mcp

Would be fun to see what your agents create.