Show HN: Ableton Live MCP | Hacker News Reader

Show HN: Ableton Live MCP

github.com

123 pointsbschoepke79 comments

Ever wanted to control Ableton with just your voice? Me too! I made this MCP server so I could just ask Codex to do anything in Ableton Live for me, while I was nap-trapped by my baby.

The chat messages I sent to Codex to make this:

in ableton, make a self reflective song, with audio vocals (via macos say) and chip tunes and 80's drum machines. should be a real edm banger

i want midi for everything but vocals please, with ableton devices. not prerendered audio for instruments

needs some fills

and should hit way harder after "3-2-1 i become the sound"

the vocals are squished too much (read too quickly), give them a little more length

add some dynamics, the song is basically one volume. and some pumping side chain

improve dynamics of the clap, seems a bit flat and indistinguished, want it harder after the 3-2-1 drop

introduce a new element on a new track after the 3-2-1 drop, that comes in but then recedes before the final exit

doesn't seem like the new thing has any notes

the element is a bit muddy/indistinct. perhaps it needs simplification and more space, different instrument choice, i dunno

Open articleView on HN

Things I would use AI for in music production:

1. Generating track layouts (add tracks + empty audio/midi clips throughout)

2. Generating MIDI sequences

3. Generating Serum patches

4. Extracting stems from existing audio

5. Automating common workflows (eg sidechaining)

6. Semantic search of sample library

That being said, I don't think I want a full agentic workflow for vibe-producing. Point solutions seems like a better fit for me, personally.

> 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.

I’ve had some fun building simple instruments in the browser using AI and piping midi to Live, then munging from there [0]. The whole principle of fully AI generated music leaves me cold but AI as a sort of sidechain to the creative process seems potentially interesting.

[0] https://variousbits.net/2026/02/22/building-generative-music...

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.

Generating MIDI sequences

https://www.muse.art/home

Interesting work and I’m really interested to see where this direction goes.

One question I keep coming back to is where tools like this MCP go beyond templating, like what I already use heavily in Ableton Live. In music production, many tasks are repetitive but not identical and that’s exactly where something like this can shine.

At the same time, music is widely seen as a "manual" craft. Every step in the chain from playing an instrument to a final music piece has both a technical and a creative side, and part of the process is staying curious and critical about what could be improved / done differently in the next project. That makes it an open question where automation actually adds value versus where it takes something away.

Where I’d personally love to see AI make a difference is in audio engineering / post processing, which also requires a lot of creativity beside a solid fundament of experience to really excel. There’s often a big gap between a great musical idea and a polished mix or master. If AI could help close that gap and contextually help to improve on things like tone, space, EQ, and loudness would be hugely valuable.

But the key for me is trust and transparency about modifications. I don’t want a black box that just makes things "better™." I’d want something that clearly explains its actions, like: "I've added an EQ to the piano on track 3 at 01:23 to open it up for the bridge so it sits better in the mix."

That kind of assistive, explainable approach would feel much more aligned with how people actually would be open for an assistant to create music.

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"

Added an "Agent Audio Tap" Max for Live device that lets the agent analyze any part of the signal chain and optimize your mix based on that (or just make cool visualizations!)

https://github.com/bschoepke/ableton-live-mcp/blob/main/READ...

Yes! I want this for MainStage -- this would allow me to automate my weekly template setup for playing at my church. Each week before practicing I look up the songs in Planning Center and create a new MainStage concert file with one patch per song, and add notes to each patch screen with the song's key, etc. Automating this would save me the time of doing the busy work and get right to practicing.

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.

We should talk! I have a cli that does exactly this, I was able to reverse engineer the MainStage concert format

> Ever wanted to control Ableton with just your voice?

Never.

I guess the guidelines don't apply to you, as long as you disagree vehemently enough with the OP's basic intent.

No amount of guidelines will make me lie in my replies.

For me, the point of making music is making it myself. If want to have something done for me I could just play someone else's record and pretend like I made it.

When we recently added MCP to Ardour (a cross-platform FLOSS DAW), the goal wasn't to get the machine to make the music for you, it was to provide alternate ways of interacting with the DAW (particularly for those with visual impairments that make voice control preferable).

The is the age-old music parochial thing. "Oh, he's just in a cover band, he doesn't write anything" / "Oh, she's just a composer, she can't even play the stuff she writes" / "Oh, he writes and plays his own stuff but knows fuck all about theory so it's not real music" / etc.

Me, I'm having a blast with claude code, MCP, and Ableton. I'm directing harmony and asking for arrangements and variations in rhythm, mixing, and production. Don't know if that counts as "making it myself", but then I was writing music before I could actually play any instrument at all, so :shrug:

I think there's a way to define it without it either side feeling personally attacked. One of the things I like about using agents for programming is that if the spec is detailed enough, I can implement it in a number of different languages and still get the thing I intended. That means that the "art" is in the spec, not the implementation.

I think the question with AI in music is when it gets to that point. What's the musical spec? What's the implementation? If the spec is supposed to be the pure distillation of my intent, then shouldn't that mean each time I engage AI to "implement" the spec, the musical output of the AI should be the same?

At that point I'm all in favor of using AI for music. But when AI is used to replace a specific intent with vague intent, that's where I feel like something is lost in the human experience of human-created music.

Previous generations might have said the same thing about Ableton itself, vs playing a physical instrument. In that regard, AI might become just another power tool for creative expression.

I’ve always said that the more divergent the input is from the resulting output, then the less personal expression you have. For me, in order of moving away from meaningful control in generative models, it goes: “text → code,” “text → picture,” and, at the very bottom, “text → music.”

For me personally, music composition begins and ends with the motif - the melody itself. It’s the part I enjoy the most, and it’s also the part I have the most individual control over since I can sing.

Everybody makes music differently, but if you lack the ability to play an instrument and you also can’t whistle or sing, it’s hard for me to imagine how you’d have any meaningful control over the melody.

How would a non‑musician express an actual melody that they came up with (beyond simple things like instrumentation and general “feelings”) in text? RED RED RED BLUE. (Sorry couldn't resist a Mission Hill reference here.)

With all that out of the way, there's still lots of room for using AI in music. I’ve used it to take some of my existing songs, mostly pianistic in nature, and swap out instrumentation and arrangements just to play around with different soundscapes. It's like BIAB on steroids.

Agree to some extent. At some point though we jump the thin line between creative expression and… magic?

Like if at some point I can just say “Generate a song similar to Smooth Criminal, different enough to not trigger copyright claims” and it just works, and everyone loves it… well is that creative thinking?

I think you can quantify the amount of creative expression you engage in by looking at all the decision points in the creative process where you are directly involved in making the decision. For an LLM prompt, that is going to be fairly limited by definition. I suppose the quality can be measured then by how novel and effective the output/approach of each decision is then, how much impact is made.

The amount of creative expression does not necessarily correlate with impact. Something can be created with nearly zero creative expression, that ends up making a significant impact. In that case you are more of a director than an artist I suppose, in that you direct the high-level process and only make decisions there. You can call it creative thinking in the same way a good businessman makes smart high-level decisions and then delegates what is downstream to others, with decisions being optimized for impact.

I think you can be creative "within a frame" in that sense, e.g. creative in the way you wield an LLM for instance, which is on a different scale compared to being creative on the piano roll with how you organize and brainstorm your melodies. It's just a different skill set at a different granularity altogether. But the one thing that I think holds, is that higher level methods have less creative expression by definition, because you are delegating more decisions to other faculties; you are seeing less of the "creator" in the work.

I think there is something to it. First, it would still need to be different enough from Smooth Criminal to avoid listeners just going back to the original. Then, if anyone could just type a simple prompt like that and get a hit, wouldn't we be flooded with 'sounds like' singles, which would turn the audience off of those, and now, you're not making hits...

I think there will always be more to it then just a simple prompt, but having the vision to make a song that sounds pleasing, and unique enough is certainly creative to me.

Of course, there's also a huge demand for generic, inoffensive music (think theme/intro songs, waiting room and elevator music). If we could make that more enjoyable to listen to, would anyone care if that's not creative thinking?

You could make (and many do) the same arguments over covers of songs, even when the covers end up eclipsing the original. Where was the creative thinking in that?

> it would still need to be different enough from Smooth Criminal to avoid listeners just going back to the original

Or just cheaper to license so that Spotify/Pandora promote it in your algorithmic feed. It's audio skimpflation!

> AI might become just another power tool for creative expression

It is NOT a digital tool to create art. Yes, people used to be snobbish about digital art. Some still are. This doesn't say anything about generative AI because that isn't a tool.

The closest equivalent is hiring someone on fiverr to create music for you and claiming you created the music because you wrote the "prompt".

There is nothing creative about using generative AI. Is is a form of management. The difference is that instead of extracting labor directly your are extracting dead labor from the million of artists whose work was stolen to train the AI.

I don't disagree if its building the whole song, but given that this is tooling within the DAW, if an artist went in and said 'give me 5 alternative reverb sounds on this track', is that not using AI as a tool? Yes, the AI is creating the sound profile, but is that any different then using presets, or samples?

I used to play around for days just making sounds on my synth. The process of creating them was often just turning random knobs and dials. If the AI is turning those for me, thats not a tool?

> I could just play someone else's record and pretend like I made it

How many "DJs" today could even find two records that they could key and beat match? Then physically mix them on the turntables with no software or sync buttons? AI is just going to make this worse...

> How many "DJs" today could even find t

A lot of them. The barriers to entry have been lowered, which also means there are way more DJs around. And some of them will start to expand their horizon.

I don't know, but I would not be surprised if the total amount of people who can mix without sync increased. Though the percentage of DJs who need sync is probably higher.

I started DJing with 'rona and now somestimes mix vinyl. And I also hosted open deck nights with CDJs where a lot of beginners did not use sync, unless they where only a few month into the game.

I don't think it's a negative.

I get why people make gut statements like this, and to me something does feel different about AI.

But I realize I have not seen any criticisms of AI generated music that are meaningfully different from criticisms I've heard of other advances/changes in music technology, whether performance or recording.

Sampling, scratching, drum machines, autotune, electric guitars even.

The main unconsidered criticism that used to come from old-school musos was that 'you press a button and the synthesizer/drum machine/whatever does it all for you'... Only now is that perhaps coming to be true.

There's a difference between technology/technique that adds a new sonic palette to the canon, and one that takes away the necessity to have any direct input in the process of production. I guess we'll find out which this is if there's a wave of novel AI assisted genres that emerge, or not, as may be the case.

Well in "traditional" music production every individual component of a song has the creative intent of the artist in it. With AI you have no idea if there is any intent or if its just something an LLM spat out.

If all you care about is the raw sound file created and you don't care about the connection you might feel with the artist behind it then maybe intent isn't relevant to you.

Each genre has a fairly tight envelope within which to operate. Regardless 90% of tracks never make it to the finish line because hobbyists haven't learnt them well enough to groove them out. If with a little help these tracks were all finished then bedroom producers will over time learn what works and be able to explore more.

I think the parent comment was saying that the problem is not quantity, but quality.

Warping my mind back into a hobby-enthusiast music producer mindset:

an MCP that generates presets for a limited pipeline with many sweet spots sounds... interesting?

To me, the idea of being able to have, say, a chain of a simple VA synth + delay + compressor and a very simple step sequencer, combined with prompting and a genAI model that spits out patches, sounds very endearing and interesting.

Much more interesting than Gemini or Suno for example.

Depends on the training and input space of course.

I deliberately described a limited setup, the controls of which could be described in less than a kilobyte.

Many dance music synth patterns could be described by simple means (tracker/step sequencer, looping, a few knobs).

That's what makes a lot of music interesting.

I can easily imagine a producer creating very individual and interesting output by unleashing the right models.

I think, just like with human producers, constraints liberate.

An AI controlling a very limited synthesis chain is more interesting than a very complex synthesis chain controlled by a human with no musical "vibe".

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.

I will caveat my first comment by also noting that I am well versed in computer music history, and read many many papers in CMJ[1] and elsewhere about generative and automatic composition tools such as Emily Howell[2]. I do NOT have a problem with generative, algorithmic and automatic composition in this sense, as an extension of the creative intentions of the human composer, in the right context. See also Autechre[3] for what can be done with Markov chains and good taste. What we are discussing here is the musical equivalent of a dishwasher.

[1] http://www.computermusicjournal.org/

[2] https://en.wikipedia.org/wiki/David_Cope#Emily_Howell

[3] http://autechre.ws/

Addendum: I would highly recommend the Margaret Boden book referenced in the wiki on David Cope/Emily Howell, which is an absolutely fascinating read and was incredibly far-sighted in its enquiries on this topic.

It'd broadly sad how folks so broadly slight and disregard novelty, are so quick to judge assume & discredit.

I have such respect for those who can do the good work of comments like your, trying to pry the closed mind open just a little more. This is such an essential outlook basis that needs to be taught, reinforced: a sense of exploring potential progress rather than sinking merely to conserving or out grouping or denying.

It's really cool that the human agency loop is improving. Ableton & DAWs should be so much better with expanded more language native interfacing!

Can I ask what the specific markers / qualifiers are for you to consider (let's call them) 'classical' generative and algorithmic techniques fair game in creative composition, but LLM agent based techniques not so?

To me, it seems like the "do it for me" aspect is similar, just at different levels of abstraction.

Firstly, they all came to the use of those techniques after having been through years of work the 'hard way', often being able to play to a conservatoire standard, and had a very extensive grounding in the tradition that came with that. Then they owned* or designed the thing they were asking to 'do it for me' and could modify it at their discretion, effectively making it an integral element of the composition. The prior training was crucial in getting anything good out of any of it IMO (high level reflection based on canon knowledge and deeply considered personal sensibility, etc.)

* I suppose in the early days, running on an mainframe would belie the definition of ownership per se, as it required access and was limited to that specific machine/institution, but then we are talking about a time where personal computing wasn't available.

Thanks for your well considered response. I disagree with the notion that extensive classical training is required in order to make beautiful, noteworthy music. There are innumerable counterproofs of this in every era of music. I also disagree that fully and deeply owning/designing one's tools is required - though I understand that we are more specifically talking about generative tools, I personally argue there's not enough meaningful distinction. One chooses to exercise intent, whether the tool is acoustic or digital, general or hyper-focused. And fully understanding the workings of every tool is a fool's errand in this modern age.

Whether these then extend to AI and LLMs I still can't fully say. There is, obviously, some kind of qualitative leap here. I'm not fully settled.

But I guess I lean more towards - it is a tool, let people use it to make their own beauty.

I didn't specify that the training had to be a classical, conservatory, background. It was only mentioned with regard to the background many of the original computer musicians came from, and which is understandable considering the era and situation of computing back then. Autechre are a good counter-example of that, which I have noted above. Two hip hop heads from the north of England, who have made some of the best contributions to electronic/computer music in recent decades. As you point out, there are loads more, not worth making a list here. Though I will still assert that I am yet to hear any good music come from someone who has anything less than a developed knowledge and passion (obsession?) for their area of interest, be that classical repertoire or drum and bass.

I wonder how one is supposed to exercise intent when the tool in question is specifically designed with the purpose of removing your ability to have direct influence on the result it produces. At best we get curation/collage, which in itself is no big change from the way things have been for decades (sample packs, premade loops, and going back further, sample CDs, for instance), but what goes away is the human touch.

The main difference is tweakability: With classical generative and algorithmic composition, the human can change parameters in real time and more closely guide the shape of the piece.

This as well. Most 'classical' algorithmic music had an element of expressiveness allowed to the composer in the moment.

> What we are discussing here is the musical equivalent of a dishwasher.

A dishwasher that may have been taught about Markov chains ...

Welcome to the era of instant gratification.

> should be a real edm banger

I'm afraid Codex ignored that one.

Very cool! I posted my own experiments in this area a few months back, which were an iteration on an existing Ableton MCP. It’s great to see more people experimenting in the spaces of interfacing with complex applications and music production.

https://news.ycombinator.com/item?id=46428922

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.

Does anyone know of other MCP servers for similar music creative tools? I'm interested in things like sonic-pi, strudel.cc and orcas. But very open to anything. I think there is a good opportunity for kids to learn using these tools, especially if I can wire it into my mycroft.ai/neon device.

MCP for Ardour was added more than a month ago, thanks to contributor zabooma:

https://github.com/Ardour/ardour/commit/d582a0b042a68ccb22c0...

I wish there was something like this for flstudio.

I've got 25 years of loops that basically to finish them need better arrangements. Using AI to auto generate sections is what I'm missing.

> I've got 25 years of loops that basically to finish them need better arrangements

Welcome to the club. You need to learn how to actually finish a track, which is the most difficult but also the most rewarding part. Why would you use AI for that? I mean, just listen to that demo track Codex made in the above repo, you surely don't want that.

There's a good book about this, published by Ableton, you can read it for free here:

https://cdn-resources.ableton.com/resources/uploads/makingmu...

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.

It's not just that the track is garbage (the "say" vocals are actually the least of its problems). Even if AI would make a good track: why use AI for creating your arrangements in the first place? Why this resistance to actually getting good at something? I can understand if your livelihood depends on it and you just need to be fast, but why for stuff you do for fun?

The book I mentioned has a good suggestion when struggling with arrangements: just copy. Take a track you really like, put it into your DAW, sync the speed and replicate its structure. You'll see that in many genres, structure is often exactly the same anyway. This can be an eye opener, and once you've realized this, you'll be able to experiment with structure in ways you couldn't do before. That's the fun part.

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.

I agree that having an MCP Ableton can make total sense. After many years of use, I would say I know Ableton quite well, but nowadays, I regularly ask ChatGPT if certain things could be done differently/more efficiently, and it often surprises me with new ways of doing things. For instance, sometimes Ableton has gotten new features over the years I'm not aware of. It surely would be nice to have this integrated via MCP.

I think you would get much better feedback if you'd focus on these use cases: flattening the learning curve for newcomers, and new ideas for experienced users, rather than creating tracks completely by AI. Because in that case, why even go through a DAW and not use Suno directly?

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.

on a side note, his early 2000s output is genius

They had the most emotion because I didn't know what I was doing. Then for any ten years I got lost trying to make it sound technically good. Now I must make stuff and there technical part happens automatically

> Welcome to the club. You need to learn how to actually finish a track

I don't think you understand. I've got thousands of songs. Why would I use Ai to generate arrangements... Maybe for ideas?

Maybe because certain things I'm lazy about?

Maybe because I've got thousands of songs?

It's not actually difficult to finish a song if your output is high enough. Sometimes the songs just come out without any struggle. But most of the time they don't.

I wrote and finished my first song around 1996. Using Cakewalk plugged into a midi keyboard.

HN is full of people who think using AI means you are lazy or can't do something. The fault is yours not mine. Adapt

Consider taking lessons from a human?

This is really cool!

This kind of automation will allow impaired people to have access to a whole new world of creation. Blind and motor impaired comes to mind.

this is awesome. Does anyone recommend one for Logic Pro X? I see a few in the wild but would love to help support if anyone is tinkering with one

Very nice! Thanks for sharing.

Thank you for sharing!

is this using M4L or the LOM ?

Object model

Careful you don't fuck up your hearing with an fx chain mistake.

If you’ve gone to the trouble of setting up Ableton MCP, you’ve already worked harder than Suno requires to make a banger

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

Suno is fun, but has absolutely nowhere near the affordances most musicians want.

Being a banger is not enough.

Suno's output is grating though?

Honestly, music generation is solved. You don’t have fine control at this point but people are unable to tell the difference anymore. There are tons of YouTube videos with blind tests between real artists and AI and people have no idea.

The Matrix wants everything solved, doesn't it?

>No alarms and no surprises

Solved !== Good

Reply on news.ycombinator.com