Blog
Read the latest news from the Chatmu blog.
Yes, there is a DAW MCP server for almost every major DAW in 2026: Ableton Live, Pro Tools, Logic Pro, Cubase, FL Studio, REAPER and Bitwig all have community projects that let Claude or another AI client control your session by text. What almost none of them do is hear the audio while you play it. They read and change tracks, clips and parameters; a few can analyze a bounce or a render after the fact.
Updated October 2026.
I'm a producer and sound engineer, and I run a label with around 500 tracks. I went through every one of these projects with the same question: if I ask "why does my chorus feel smaller than the verse?", can the agent answer from what it hears, or only from what it can read in the session? This guide maps the options per DAW, links only repos I checked on GitHub, and shows where each one stops.
MCP (Model Context Protocol) is the standard that lets an AI client like Claude, ChatGPT or Cursor call tools on your computer. A DAW MCP server translates those tool calls into commands your DAW understands. If you're new to MCP, start with our complete guide to MCP for the music industry.
Every DAW exposes a different door, so every MCP server is built differently:
All of these doors carry commands and session state. None of them carries the audio signal by default. That is the core limitation of most DAW MCP setups.
| DAW | MCP options (GitHub) | Controls session | Hears audio | Setup |
|---|---|---|---|---|
| Ableton Live | ableton-mcp (ahujasid), LivePilot (dreamrec), Anthropic's Ableton connector | Yes (community MCPs); the Anthropic connector answers from the manual | ableton-mcp: no. LivePilot: spectral data through an optional Max for Live device on the master, plus analysis of rendered audio | Python/uv + Remote Script; LivePilot adds a setup wizard |
| Pro Tools | protools-mcp-server (skrul), protools-mcp (BlueElevatorProductions) | Yes, via PTSL | skrul: bounces a region and analyzes the file (peak/RMS, waveform and spectrum images). BlueElevator: no | Node or Python 3.11, PTSL enabled, macOS for BlueElevator |
| Logic Pro | logic-pro-mcp (MongLong0214), logic-pro-mcp (koltyj) | Yes, through macOS automation | No live listening | Homebrew or Swift build; Accessibility and Automation permissions |
| Cubase | cubase-mcp (hedidjs), yoka-cubase-mcp (yokadeeds-dev) | Partly: transport, mixer, commands over MIDI | No | Python + virtual MIDI port + Cubase remote setup |
| FL Studio | fl-studio-mcp (karl-andres), flstudio-mcp (rosasynthesiz) | Yes, within FL's scripting limits | No live listening; rosasynthesiz analyzes audio files you export manually | Python + loopMIDI (Windows) or IAC Driver (macOS) + controller script |
| REAPER | reaper-mcp (itsuzef), total-reaper-mcp (shiehn) | Yes, deep | itsuzef: renders the project to a temp file, then measures loudness, spectrum, dynamics, stereo | Python 3.10+ (pip) + distant API; shiehn uses a Lua bridge |
| Bitwig | bitwig-mcp-server (WeModulate), daw-mcp (ptaczek) | Yes: transport, mixer, device parameters | No | Python/uv + DrivenByMoss OSC; README says work in progress |
| Ableton Live (from Oct 31) | Chatmu Desktop | Yes | Yes: the Coproducer plugin on the master streams the mix when you hit play | Desktop app for Mac and Windows |
Star counts and features are from each repo's GitHub page in October 2026. These are open-source projects; check each README before installing.
ahujasid/ableton-mcp is the most popular DAW MCP on GitHub (over 3,000 stars). It installs a MIDI Remote Script that opens a socket inside Live, and a Python server (run with uv) that Claude talks to. You can create tracks, write MIDI clips, load instruments and set tempo by typing. It does not hear the audio.
dreamrec/LivePilot goes further: a Remote Script, a large tool set, and an optional Max for Live analyzer on the master track that sends spectral data to the server, plus analysis of rendered audio. The 38 spectral tools require that bridge. If you have Max for Live and are comfortable with a Python setup, it is one of the most complete open projects for Live.
Anthropic's Ableton connector for Claude, launched in April 2026 as part of Claude for Creative Work, grounds Claude's answers in the official Live and Push documentation. It is a good tutor: ask how sidechain routing works in Live 12 and you get an answer from the manual instead of a guess. It is built to answer questions, not to listen to your set.
There are two serious community options for Pro Tools, both built on PTSL. I did not find an official MCP server from Avid.
skrul/protools-mcp-server is a Node server that runs read-only by default with granular write permissions. Its README is honest: AI models can't listen to audio directly, so its analyze tool bounces a region (one track or the full mix) and returns waveform and spectrum images, peak/RMS and onset data. It needs the PTSL proto file from Avid's SDK.
BlueElevatorProductions/protools-mcp is a Python 3.11 server for macOS with 25 tools, built for podcast post: session info, transcript search, markers, editing and AAF export.
Logic has no public scripting API, so its MCP servers automate the app. MongLong0214/logic-pro-mcp is a Swift server installed with Homebrew, validated on Logic Pro 12.3, with confirmation before destructive actions. koltyj/logic-pro-mcp combines CoreMIDI, Accessibility, CGEvent, AppleScript and OSC. Both need macOS Accessibility and Automation permissions, and you'll re-grant them after upgrades. Neither listens to your mix while it plays.
Cubase is the thinnest category. hedidjs/cubase-mcp is a Python server that drives Cubase through a virtual MIDI port and Generic Remote: transport (play, stop, record) and mixer (volume, pan, mute, solo). yokadeeds-dev/yoka-cubase-mcp uses the Mackie Control protocol and reads state back, with a README in German. Both are small projects. If you search "cubase mcp" expecting Ableton-level control, expect transport and mixer, not composition or listening.
karl-andres/fl-studio-mcp (around 240 stars) covers transport, mixer, channel rack and piano roll over MIDI, with installers for Mac and Windows. Its README notes FL's scripting API can't load new plugins or create patterns. rosasynthesiz/flstudio-mcp adds 67 tools including mixing helpers; because FL's API can't render audio, you export the file yourself and Claude analyzes it afterwards (tempo, key).
REAPER is the most scriptable DAW, and it shows. itsuzef/reaper-mcp (pip install, 61 tools) covers tracks, MIDI, FX, rendering and analysis; its analysis tools render the project to a temporary file and then measure loudness, spectrum, dynamics and stereo field. shiehn/total-reaper-mcp aims for full ReaScript coverage through a Lua bridge.
WeModulate/bitwig-mcp-server controls transport, mixer and device parameters over OSC through DrivenByMoss; its README marks it as a work in progress. ptaczek/daw-mcp targets both Bitwig and Ableton.
A session file tells you a fader is at -6 dB. It doesn't tell you the bass and kick are fighting at 80 Hz, that the vocal is buried in the chorus, or that the master hits -7 LUFS with true peaks over 0 dBTP. Mix decisions are made with ears:
An agent that only reads parameters can guess at these. An agent that analyzes a bounce can measure them, but only after an export, and only for that one render. The step that's missing is listening to what the DAW is playing right now, while you work.
Chatmu is an AI agent for the music industry with its own catalog of music-industry MCP connectors. Tomorrow, October 31, we launch Chatmu Desktop, for Mac (Apple silicon and Intel) and Windows, starting with Ableton Live 11 and 12. More DAWs are coming.
It hears your session. The Chatmu Coproducer plugin sits on Ableton's Master and streams the mix to the app when you hit play. Chatmu Ears measures it: loudness (LUFS, true peak), spectrum, dynamics, stereo image, defects like sibilance, mud and masking, null tests, and loudness-matched A/B comparisons. Every number comes with a unit and a target, and the chat shows a spectrogram of what it heard.
It acts on your set. Create, name, arm and record tracks; move faders, pan and sends; set plugin and device parameters in real units; write MIDI clips; manage scenes, locators, tempo and sidechain; group tracks; export the song, a range or up to 32 stems. Reading is free; writing needs "Allow control", and destructive actions ask for approval every time.
A typical prompt: "Listen to the chorus and tell me why it feels smaller than the verse." Chatmu plays it, measures it, and answers with numbers (for example, the chorus is quieter and narrower than the verse, and the bass masks the vocal in a specific band), then proposes the fader and EQ moves and makes them once you allow control.
It is not music made by AI. It's AI inside your DAW: you decide, everything stays editable, the stems are yours, and it sounds like you. When the mix is ready, the same agent can take the export to distribution, video and social. If you already use Chatmu inside Claude or ChatGPT, see Chatmu MCP. There's a free plan; see pricing.
Yes, community ones. skrul/protools-mcp-server and BlueElevatorProductions/protools-mcp both connect to Pro Tools through PTSL. I did not find an official MCP server from Avid.
Yes. Community servers like ahujasid/ableton-mcp and LivePilot let Claude create tracks, clips and parameters through a Remote Script. Anthropic's own Ableton connector answers questions from the Live and Push manuals.
There are small community projects (hedidjs/cubase-mcp, yoka-cubase-mcp) that control transport and mixer over MIDI. None of them listens to the audio.
Most DAW MCPs can't. Some analyze a bounce or render after you export. Chatmu Desktop listens to the Ableton master when you hit play, through the Coproducer plugin and Chatmu Ears.
AI for the industry. Humans for the music.
