bschoepke

> For (1) and (2), the worst-case just doesn't occur. I mean, installing packages and type checking can surely be slow. But, at least in my career, I've never seen a galactic blow-up.

Hehe, clearly the author hasn't written any SwiftUI.

Rust certainly helps for certain classes of errors but doesn't help at all for logic errors or incompletely understood requirements or occasional need for hacks due to business needs.

If you really want, you can cmd+T and go to chatgpt.com within the ChatGPT app haha

Thank you for the feedback, I'll record some more co-creating examples! And yea, it's also fun to take stuff out of Ableton and run it through Suno, to get real vocals and such.

A cool thing about this MCP is you can ask Codex Ableton questions and it will go and read the state of your current Live Set and answer based on that. You don't have to have it change anything for you if you don't want.

> 6. Semantic search of sample library

I just pushed another tool for that! It wasn't exposed in the Live API but the implementation just issues the same queries to the underlying sqlite db that the Live GUI queries for the "Find Similar Sounds" feature.

Well, I do that too. I actually spend most of my music hobby time with my Ableton Move. I just happen to find it very fun to play with Codex as well.

Well, it's also interesting to take a rough sketch out of Ableton, render it to audio, and then use Suno to "upscale" it.

Nice, the audio analysis M4L device is something I wanted to add next, as that would close the feedback loop and let the agent actually analyze its setup. Kind of like how giving agents screenshots of its UI work helps improve the UI.

I agree! I made this mostly so I could stay in the flow more in Ableton without getting lost spinning knobs and pressing buttons. Although I do enjoy spinning knobs and pressing buttons.

With this I can keep my hands on my keyboard or guitar and direct Codex to make a quick backing track.

To be fair, the demo track was one of the first I had it make, and I didn't put much effort into it because I thought it was especially funny with the macos "say" command vocals.

It's a garbage-in, garbage-out situation. If you give it more musical direction you will get more out of it.

Interesting use case! I'm not familiar with MainStage, but that workflow is possible with Ableton using this MCP. Codex itself can pull info from websites or other apps using its existing toolset and then push it into Ableton using this MCP.

Same here! I tried all of that, have 1, 2, and 5 working. So far it doesn't seem like Ableton's stem splitter or semantic search are programmatically accessible, but I didn't try very hard. I do have Serum so maybe I'll look into its file format; that does seem doable. The MCP already enables the agent to make patches for built-in Ableton devices.

Thank you, this gave me an interesting idea: while it can already set up a mastering chain for you, it is missing the full feedback loop of actually analyzing the processed audio signal. I think I'll add a small Max4Live device that enables the agent to probe the signal chain wherever it wants; from there it can write python to analyze the audio.

And yea, I kept the MCP general purpose so you can use it however you want. You can use it to simply ask questions about your live set; it can see the arrangement, see all the settings of all your devices, see all the MIDI, etc. I had it "fix a discordant sounding section" of one of my songs; it was like "oh yea tracks 3, 5, 6 all had some unfortunate notes colliding"