Ableton Live MCP server: what an agent can actually do
Published September 2026 · Updated October 2026
Doseedo runs a hosted MCP server at https://api.doseedo.com/mcp that lets an AI agent build, convert, read and download Ableton Live projects: turn a mix into a native .als, move a set between ten DAW formats, restructure an arrangement through REAPER's plain-text format, hand you back a project you open in Live. What it does not do is drive the Live set that's open on your screen right now. No clip launching, no realtime device knobs. If that's what you came for, the answer is AbletonOSC or pylive, and this page will point you there rather than blur the line.
Two different things called "Ableton MCP"
"Ableton MCP" is the most-searched of these phrases and the most ambiguous, because the best-known project under that name does the opposite of this one:
- Realtime control of a running set. Live has a genuine internal API (the Live Object Model), and community bridges expose it: AbletonOSC (a control-surface script speaking OSC) and pylive (a Python wrapper over it), plus Max for Live from inside the set. These fire clips, set device parameters and read transport state while Live is open. Nothing here replaces them.
- Agent access to the project itself. Reading what's in a set, generating a new one, converting it, restructuring it, downloading it. That's a file-level problem: it works with Live closed, and on machines without Live installed. It is what the Doseedo MCP server does.
The two don't compete; they don't even touch the same objects. AbletonOSC never opens an .als on disk; a file-level tool never sees your playhead. Use both: prepare the set with one, perform it with the other.
Setup
claude mcp add --transport http --scope user doseedo https://api.doseedo.com/mcp
That line is for Claude Code; the first tool call opens a browser to sign in and approve. Headless clients append --header "X-API-Key: dsk_live_..." with a key from Settings → API keys. The same URL installs in ChatGPT, Cursor, Gemini CLI, OpenAI Codex, VS Code and the Claude phone and web apps, one page per client. Nothing to self-host; the reference lists every tool.
The seventeen directions doseedo_convert_project accepts for a Live set, straight from the tool's own enum: logic2ableton ableton2logic fl2ableton ableton2fl reaper2ableton ableton2reaper protools2ableton ableton2protools cubase2ableton ableton2cubase dorico2ableton ableton2dorico garageband2ableton bitwig2ableton ableton2bitwig studioone2ableton ableton2studioone, plus auto2ableton when the agent should detect the source.
What the agent can do with an Ableton project
You talk to it in project terms; it handles uploads, polling and downloads itself:
> Here's a stereo bounce of a demo. Build me an Ableton set
from it, with MIDI for the bass and a chord guide track.
> Convert ~/Music/Sets/BridgeIdea.zip (an Ableton set) to
Logic, and tell me what won't survive the trip.
> This set has 40 unnamed tracks. Convert it to REAPER's text
format so you can rename them from what's on each one, then
convert it back to .als.
Concretely, the capabilities that apply to Live:
- Build a set from audio. A full mix in, a native
.alsout: separated stems on named tracks, a detected tempo map, optional per-stem MIDI and a chord guide. About four minutes. See audio to session. - Convert, natively. All 75 directions across ten DAW formats: Logic Pro, Ableton Live, FL Studio, Pro Tools, REAPER, Cubase, Dorico, Studio One, Bitwig and GarageBand. Tracks, MIDI, audio, tempo and markers arrive editable; on the Logic, Ableton, Pro Tools and REAPER pairs automation, buses/sends and plugin state carry too.
- Read a set. An
.alsis gzipped XML, so an agent can unzip one and answer "what's in here?" with Live closed. The browser .als reader shows the same on a page: tempo, tracks, devices, plugins and referenced audio. - Restructure from code. Convert to REAPER's plain-text
.rpp, edit the arrangement as text, convert back to.als. Structural work survives; a week of automation does not; see what carries over. - Download the result as a native project file you open in your DAW.
Long jobs are async: the submit tool returns a task_id and a wait tool polls it until the artifact is ready. A session build from a full mix takes about four minutes; a conversion is quicker.
Your audio doesn't necessarily leave the machine
For files already on your disk, the agent passes the local path: the cloud builds the session structure (tracks, regions, buses, routing), and the download hands back copy commands that place your own files into the project. Audio uploads only when the server needs the samples: separation, transcription, conversion, generation or a bounce.
Install it and try one conversion
Add the server, compress a Live set folder, and ask your agent for a Logic version plus the list of what did not survive. The report is the part a drag-and-drop converter never gives you.
The Doseedo MCP server →What it does not do: read this before you install
- No live control of a running Live set. No transport, no clip launching, no device automation on the open set. Use AbletonOSC or pylive; the map is in controlling Ableton Live with Python.
- Writing edits into a project on your Mac targets Logic Pro today. That half of this product runs through the Doseedo Desktop bridge on macOS, which saves, closes and reopens the Logic project to write each batch of edits in.
- Offline bounce is Logic-side too. Printing tracks through their real third-party plugin chains with the DAW closed runs through that same bridge.
- It is not a replacement for Live. You still mix and arrange in the DAW; this moves projects in and out of it.
- Automation and plugin state have limits off the direct Logic↔Ableton path. Restructure arrangements this way, not finished mixes.
Why Logic gets the desktop bridge and Live doesn't
It's the inverse of what the APIs suggest. Logic ships no public realtime scripting surface (its Scripter plugin is JavaScript confined to one channel strip), so Doseedo reaches a Logic project through a native app instead: the desktop app saves it, closes it, writes the edits into the file and reopens it (the Logic side, in detail). Live already has the Live Object Model and a mature stack around it, so the gap worth closing there was never realtime control: it was the project file: generating sets, converting them, restructuring one without hand-editing XML Live will refuse to open.
Which tool for which job
- Fire clips or turn knobs on a set that's open right now: AbletonOSC or pylive.
- Ship control logic inside the set: Max for Live.
- Generate, convert, restructure or download projects from a phone, a server, or a machine without Live: the MCP server.
- Have an agent's edits written into a project on your Mac: Logic Pro, with Doseedo Desktop running; it closes and reopens the project to apply them.
FAQ
Is there an MCP server for Ableton Live?
Yes, for project files. Doseedo runs a hosted MCP server at api.doseedo.com/mcp that lets an agent build, convert, read and download Ableton Live projects, including turning a mix into a native .als and converting a Live set to or from Logic Pro, FL Studio, Pro Tools and REAPER. It does not connect to the Live application running on your screen.
Can Claude or ChatGPT control Ableton Live in realtime?
Not through this server. Firing clips, moving faders and turning device knobs on a set that is currently open is what AbletonOSC and pylive do, via Live's own Live Object Model. Those are realtime Python tools running on the machine with Live open; they do not read or write project files, and this MCP server does not replace them.
What can an AI agent do with an .als file?
Build one from a mix (separated stems on named tracks, tempo map, optional per-stem MIDI and a chord guide), convert one to or from Logic Pro, FL Studio, Pro Tools and REAPER, restructure one through REAPER's plain-text format and convert back, and download the result as a native project you open in Live.
Can it put the stems of a mix straight into a Live set?
Yes. That is the one build that targets Live directly. audio_to_session with target:"ableton" returns a native .als with the separated stems on named tracks, a tempo map, section markers and, if asked, per-stem MIDI and a chord guide; separate_stems with export:"ableton" does the lighter version. Writing edits into a project on your Mac is Logic Pro only, through the macOS desktop bridge, which closes and reopens the project to apply them; a Live set is handed back as a file.
Do I have to upload my audio?
Not always. When a session is built from files already on your disk, the agent passes the local paths and the cloud builds the arrangement (tracks, regions, buses and routing), then the download hands back copy commands that place your files into the project. Audio uploads only when the server genuinely needs the samples: separation, transcription, conversion, generation or a bounce.
How do I install it in Claude, ChatGPT or Cursor?
In Claude Code, run claude mcp add --transport http --scope user doseedo https://api.doseedo.com/mcp. The first tool call opens a browser for OAuth. In the Claude phone and web apps it installs as a custom connector with the same URL, in ChatGPT as a custom MCP server, and any MCP client can register it as a Streamable HTTP server, optionally sending an X-API-Key header instead of using OAuth.
The Doseedo MCP server · one URL, every client
- Any AI agentwhat works today
- Claude & Claude Codephone, desktop, terminal
- ChatGPTas a custom MCP server
- Cursormcp.json, one entry
- Gemini CLIgemini mcp add
- OpenAI Codexcodex mcp add
- VS Code Copilotmcp.json, type http
- All 33 toolsthe reference
- Convert projectsten DAW formats, 75 directions
- Separate stemsvocals, drums, bass, more
- Audio to MIDItranscription, chords, key
- Generate musicsongs, parts, covers
- Bounce offlinethrough real plugins
- Swap a drum packaudio stays local