Ableton Mcp Extended

AbletonMCP — Extended Fork (Measure · Set · Generate · Auto-fix)

> This is a fork of ahujasid/ableton-mcp (MIT, © Siddharth Ahuja). > It keeps everything the original does and adds device-parameter read/write, > offline audio analysis, single-cycle wavetable export, and an autonomous > anti-masking chain. All credit for the original architecture goes to the > upstream project — please ⭐ it.

AbletonMCP connects Ableton Live to an MCP client (Claude Desktop / Cursor) so the model can control a live set. The stock version can build and trigger things (tracks, clips, notes, load devices). This fork additionally lets the model hear, set, generate and self-correct — closing the loop measure → set → re-measure.

---

What this fork adds

| Area | Stock | This fork | |---|---|---| | Device parameters | load only | read (get_device_parameters) + write (set_device_parameter) + delete (delete_device) | | Audio analysis | — | analyze_audio (peak/RMS, bands, LUFS, time-window), find_frequency_masking, analyze_mix | | File formats | WAV/AIFF/FLAC | + m4a/mp3 (via audioread + ffmpeg) | | Generate | — | extract_single_cycle (single-cycle wavetable WAV) | | Audio→file bridge | — | get_clip_sample_path, list_audio_clips | | Mix automation | — | auto_fix_masking — pulls clips + mapping straight from Live, sets EQ-Eight high-pass against masking | | Sidechain | — | configure_sidechain (sets compressor values; the sidechain switch stays manual — Live API limit) | | Render/capture | — | capture_track_to_wav (real-time resampling) | | Stability | get_track_info could throw a C++ TPyHandle error | reads run thread-safely on Live's main thread + fresh song handle |

⚠️ Experimental (not yet confirmed in a live Ableton session)

capture_track_to_wav, delete_device, and the get_track_info thread-safety fix are implemented and import-clean, but the live-operation behaviour still needs real-world testing. The offline analysis/generation tools are verified.

Honest API limitations (not bugs — unavailable in the Live Object Model)

• No offline render/bounce/freeze → capture_track_to_wav uses real-time resampling; manual File > Export Audio remains the pragmatic fallback.
• No device reorder (only delete) → load devices in the order you want (they append to the chain end).
• Compressor Sidechain switch + "Audio From" source can't be set.
• Wavetable custom-file import can't be triggered (drag it in manually).
• VST/AU "Configure" can't be clicked to expose parameters.

---

Requirements

• Ableton Live 11+ (developed against 11.3.x)
• Python 3.10+
• uv (brew install uv on macOS)
• ffmpeg on PATH (only needed for m4a/mp3 analysis): brew install ffmpeg

Python deps (resolved automatically by uv from pyproject.toml): mcp[cli], numpy, soundfile, audioread, pyloudnorm (pulls scipy).

---

Install

1. Remote Script (inside Ableton)

Copy AbletonMCP_Remote_Script/__init__.py into a folder named AbletonMCP in Ableton's user remote-scripts location, then fully restart Live:

• macOS (reliable): ~/Music/Ableton/User Library/Remote Scripts/AbletonMCP/__init__.py
• Windows: %USERPROFILE%\Documents\Ableton\User Library\Remote Scripts\AbletonMCP\__init__.py

Then in Live: Settings → Link, Tempo & MIDI → Control Surface → AbletonMCP, Input/Output = None.

2. MCP server (Claude Desktop)

Edit claude_desktop_config.json. To run this local fork (so your changes take effect), point uv at the repo:

{
  "mcpServers": {
    "AbletonMCP": {
      "command": "uv",
      "args": ["run", "--directory", "/ABSOLUTE/PATH/TO/ableton-mcp", "ableton-mcp"]
    }
  }
}

(Use the absolute path to wherever you cloned this repo. On macOS, command may need the full uv path, e.g. /opt/homebrew/bin/uv.)

Restart Claude Desktop — you should see the AbletonMCP tools (hammer icon).

---

Typical workflow

1. Have audio clips in the set (or Freeze+Flatten instrument tracks first).
2. analyze_mix (no args) → pulls clips + builds the mapping from Live, ranks masking conflicts.
3. auto_fix_masking (dry run) → review the plan → run again with apply=true.
4. capture_track_to_wav or manual export → analyze_mix again to verify.
5. Always trust your ears — the recommendations are starting points.

---

Privacy / Telemetry

The upstream project includes optional anonymous telemetry. In this fork it is off by default: consent defaults to False, and no MCP_Server/config.py (which held the backend credentials) is shipped, so nothing can be sent. It only activates if you add your own config and set ABLETON_MCP_TELEMETRY_CONSENT=true. Hard-disable anytime with DISABLE_TELEMETRY=true.

---

License & attribution

MIT — see [LICENSE](LICENSE). Original work © 2025 Siddharth Ahuja (ahujasid/ableton-mcp). This fork preserves the original license and adds the features listed above.

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: motodigitalguru-beep
• Source: motodigitalguru-beep/ableton-mcp-extended
• License: MIT

Install and usage instructions live in the source repository linked above.