Clipto MCP - Let agents source clips from terabytes of your local video

Clipto MCP gives Claude, ChatGPT, and other AI agents the ability to source clips and more from inside the videos, photos, and audio recordings stored on your computer. Instead of manually browsing files, simply describe what you need. For example, turn a script into a video by matching each sentence with your local footage; find every scene where someone mentioned a topic; create rough cuts; or search years of media as if you have a dedicated assistant editor.

Add a comment

Replies

Best

Congrats on the launch! Going fully local is the right call at this scale, since nobody's realistically uploading 2TB of footage to the cloud just to make it searchable. Smart to lead with the indexing benchmark too, because slow first-time indexing is usually what kills local AI tools. How big does the index itself end up for a 2TB library, and does search stay snappy as the library grows?

Thanks Lisa! Great questions. On indexing speed, as a reference, we can process around 4TB in 24 hours on an M5 Pro, so a 2TB library would take roughly half that time, depending on the device and media mix. The index size depends on the number of files and the size/type of the source media. For a typical 2TB library, the index is around 18% of the original media size (~368GB). For predominantly 4K/8K footage, that ratio drops to around 8% (~163GB). In our current testing, libraries of up to ~2,000 video files return search results within 5 seconds.

The Elon/Daft Punk demo is wild lol.

How much manual cleanup did you actually skip vs what Claude + FFmpeg pulled off on its own?

 Thanks Boyuan! ;) actually that's the point, we left everything to Claude, the video was produced with zero-human-touch. We are preparing the Elon Musk's clips so you can give it a try.

Check back here in a couple of hours.

This might be one of the more practical MCP use cases I’ve seen for creators.

 Candy, thanks for the nice words! It means a lot for us.

The MCP integration is a smart move. Your existing AI tools can basically talk to your Clipto library directly.

 Exactly. That’s how we think about it too. Clipto becomes the media understanding layer, while you keep using the AI tools you already work with. Thanks William!

That’s the idea — one memory, accessible to whichever AI tools you choose to use.

The editing example got me thinking about how much time is wasted just hunting for the right shot before the actual work even starts.

 That’s such a real part of editing. Finding the right shot can take longer than actually using it, especially when you’re working across years of footage. We want Clipto MCP to handle that search so you can spend more time on the edit itself. Thanks!

Congrats on the second launch. That music video went out as the raw first output, no cleanup at all. Takes some nerve to show the actual thing rather than the polished version.

 Exactly. We wanted to show what Clipto MCP and an AI agent can actually produce on their own today: understand the creative direction, find the right moments across a large media library, and assemble them into an editable rough cut. A polished version would have looked better, but it would also have hidden where the product truly is. What matters most here is being able to pinpoint the exact moments you need across huge, complex libraries and turn them into a useful starting point for editors.Thanks Lucas!

 Thanks Lucas! We think the best way to build trust is simply to show what actually happened. We even shared the original source videos and the exact prompt we used, so anyone can try to reproduce the result themselves.

One thing we’re increasingly convinced of: AI doesn’t need to replace the editor to be useful. If it can turn thousands of hours of footage into a surprisingly good first cut, the editor gets to start with taste and judgment instead of search and assembly. That’s already a pretty big shift.

We had a good laugh when we saw the first output 😂 Hope you enjoyed it too!
Super exciting launch. Congratulations! The idea of AI agents directly tapping into TB‑level local media feels like a real step forward. How do you see creators using this for everyday workflows?

 Thanks! It starts by downloading the app, adding your media, and giving Clipto some time to index the library. The initial indexing only needs to happen once, and Clipto will analyze new media as you add it.

Once that’s done, you can search your library smoothly inside the Clipto app, or connect it to your preferred AI Agent through MCP for more advanced workflows. You can find a specific shot, gather useful moments around a topic, match B-roll to a script, or locate an exact line in a long recording. The goal is simple: less time hunting for footage, and more time creating.

One-shot editing is a big one! Imagine giving your agent a script and having it find the right moments across TBs of footage and put together a first cut for you. Our Elon experiment is such a case.

This makes Clipto feel more like infrastructure behind your AI tools than another standalone productivity app.

 That’s the idea behind the memory layer. We’re not trying to build another Agent or compete for where people spend their time. We want Clipto to be the infrastructure underneath the tools they already use—giving those Agents useful memory and context from their local media. Thanks!

Love the local-first approach! Curious — how does it handle matching accuracy when searching dialogue vs. visual scenes across years of footage? Congrats on the launch! 🎉

Thanks Christine! Great question. We index dialogue and visual scenes separately — transcripts and speakers for what was said, and scene-level understanding for what was shown — then bring those signals together for retrieval. So you can search by dialogue, visuals, or both, even across years of footage. Thanks for the support! 🙌

Congrats on the launch! Local-first plus MCP is a smart combination. Being able to point an agent at years of footage without uploading any of it solves a real problem.

Thank you! You nailed it — the idea is to make years of personal media accessible and useful to AI agents without requiring everything to be uploaded to the cloud first. That’s exactly what we’re building with our local-first approach. Really appreciate the support!
First
Previous
•••
456