Ableton for AI

Ableton for AI

[!IMPORTANT]

For AI Agents: Please read AGENTS.md for detailed technical context and protocol specifications before working on this

project.

Ableton for AI is the bridge between your DAW and AI models. It makes Ableton projects hearable and visible to AI models by implementing the Model Context Protocol (MCP) and exporting detailed project data.

What can it do?

Connect Ableton Live directly to MCP-capable AI clients (like Claude Desktop, Cursor, or Cline). The AI can inspect, analyze, and even modify your project in real-time.

Included Tools:

• get_overview: SESSION DISCOVERY. Provides tempo, locators, and a list of all tracks with their current mixer state (volume, panning, etc.).
• get_track: TRACK INSPECTION. Comprehensive data for a single track: metadata, complete device chain, and all device parameters (with UI-readable strings).
• get_tracks: BULK TRACK INSPECTION. Query full data for a range of tracks in a single parallelized call.
• analyze_stems: DEEP AUDIO ANALYSIS. Triggers generation of audio summaries (.summary.json) and spectrograms (.spectrogram.webp) for all tracks.
• set_track_volume / set_track_panning: MIXER CONTROL. Remotely adjust track faders and panning.
• set_device_parameter: REMOTE CONTROL. Precisely adjust any parameter of any device in the project.

Included Resources:

• ableton://stems/{track_name}/summary: AUDIO SUMMARY. Compact analysis (JSON) containing LUFS, Peak, RMS, 10ms-frames, detected transients, and a legend field.
• ableton://stems/{track_name}/spectrogram: VISUAL ANALYSIS. Log-frequency spectrogram image (WebP) for visual frequency inspection.
• ableton://stems/available/summaries: Discovery resource for all tracks with available audio summaries.
• ableton://stems/available/spectrograms: Discovery resource for all tracks with available spectrograms.

---

How do I get started?

1. Prepare your environment

Install AbletonOSC

This project relies on AbletonOSC to communicate with Ableton Live.

• Download and install AbletonOSC.
• Follow the instructions there to add it as a Control Surface in Ableton Live's Link/Tempo/MIDI settings.

Install ableton-for-ai

Option A: uv (Recommended)

uv lets you run the server directly without installing anything globally:

brew install uv          # macOS (Homebrew)

or: curl -LsSf https://astral.sh/uv/install.sh | sh (macOS/Linux)

or: powershell -c "irm https://astral.sh/uv/install.ps1 | iex" (Windows)

Option B: pip

Requires Python 3.11+ (python.org or brew install python):

pip install ableton-for-ai

2. Connect to your Chatbot (MCP Client)

Add the server to your MCP client configuration (e.g. Claude Desktop, Cursor, or Cline).

Via uvx (if you installed uv)

{

"mcpServers": { "ableton-for-ai": { "command": "uvx", "args": ["ableton-for-ai"], "env": { "STEMS_SOURCE_DIR": "/path/to/your/exported/stems" } } } }

Via pip (if you installed with pip)

{

"mcpServers": { "ableton-for-ai": { "command": "ableton-for-ai", "env": { "STEMS_SOURCE_DIR": "/path/to/your/exported/stems" } } } }

[!NOTE]

`STEMS_SOURCE_DIR` must point to the folder where your exported audio stems are located. You need to export stems manually from Ableton Live (e.g. via "Export Audio/Video" with individual tracks selected), as there is currently no OSC API for triggering exports. The audio analysis tools (analyze_stems) will look for files in this directory.

Where to put this config

• Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
• Claude Desktop (Windows): %APPDATA%\Claude\claude_desktop_config.json
• Cursor / Cline: See their respective MCP documentation.

Restart your MCP client after saving the config.

3. Use it

Once connected, you can ask the AI about your project state. The AI will use the included tools and resources to fetch data directly from Ableton.

For audio analysis: Export your stems from Ableton Live first (e.g. via *File → Export Audio/Video* with individual tracks selected) into your configured STEMS_SOURCE_DIR. This step must be done manually, as there is no API for triggering exports.

For complex tasks (e.g. mix feedback): Give the chatbot context about your project. A good starting prompt includes genre, goal, and where your exported stems start (beat or locator). See agent-templates/mix-engineer.md for a ready-to-use prompt template.

Example prompts:

• "What is the current tempo and how many tracks do I have?"
• "Check the EQ settings on my Lead Synth track."
• "Show me an overview of the mixer state."
• "Analyze the stems and tell me if there are any frequency clashes."
• "Show me the spectrogram for the 'Kick' track."

---

⚙️ Configuration

You can customize the server behavior by setting environment variables or editing config.py (when running from source).

---

🔌 Using Custom VSTs / Plugins

1. Register the Plugin Name

In config.py, add the exact name of the plugin as it appears in Ableton to RELEVANT_DEVICE_NAMES.

2. Auto-Populate Parameters (The Options.txt Trick)

By default, Ableton Live only exposes the parameters of a third-party plugin, if the overall parameter number threshold (_PluginAutoPopulateThreshold) is below 64 ( default), see here. You can change this to 128 and hope that your favorite custom VSTs will be exposed. If they are still not exposed, because there are too many (e.g. Fab Filter Pro Q 4), you have to configure the parameters within ableton.

Force Ableton to automatically expose parameters by adding -_PluginAutoPopulateThreshold=128 to your Options.txt:

• Mac: ~/Library/Preferences/Ableton/Live [Version]/Options.txt
• Windows: %AppData%\Ableton\Live [Version]\Preferences\Options.txt

Mac Terminal Commands

To quickly set this up on a Mac, you can use these commands (replace [Version] with your actual Live version, e.g., 12.3.6):

# 1. Navigate to the Preferences folder

cd ~/Library/Preferences/Ableton/Live\ 12.3.6/

2. Create the Options.txt file (if it doesn't exist)

touch Options.txt

3. Append the setting to the end of the file

echo "-_PluginAutoPopulateThreshold=128" >> Options.txt

OR open the file in a text editor and add the setting manually, e.g.

idea ~/Library/Preferences/Ableton/Live\ 12.3.6/Options.txt

---

📂 CLI: Analyze & Upload (No MCP required)

This method is ideal for LLMs that don't support MCP directly (like ChatGPT or the Claude Web Interface). It runs the same analysis pipeline locally and exports the results as files you can upload to any chatbot.

Available Commands

Full audio analysis (summaries + spectrograms + full analysis JSONs):

ableton-for-ai-cli analyze_stems

Extract Ableton project metadata only (tracks, devices, parameters):

ableton-for-ai-cli extract_ableton_project_data

Full pipeline (audio analysis AND project data extraction):

ableton-for-ai-cli analyze_stems_and_extract_ableton_project_data

[!NOTE]

The OSC daemon starts automatically — no need to launch it manually.

Output

The results will be stored in the ./out directory, organized into subfolders:

• ./out/project/: Contains the Ableton project JSON file.
• ./out/summaries/: Contains compressed audio analysis JSONs (optimized for AI).
• ./out/spectrograms/: Contains spectrogram WebP images for each stem.
• ./out/analyses/: Contains full high-resolution analysis JSONs.

Mixing with AI (Manual Upload)

1. Export your stems from Ableton into your STEMS_SOURCE_DIR.
2. Run ableton-for-ai-cli analyze_stems to generate analysis files.
3. Drag and drop the files from your out folder into the chatbot (spectrograms .webp + analysis .json).
4. Use a prompt like this:

I have attached several audio analysis files (spectrograms and energy analysis).

The given files only reflect a part of the project timeline, see below in Part.

Genre: Drum and Bass Goal: Improve the clarity of the lead synth and make sure it doesn't clash with the vocals. Part: The *.analysis.json and *.spectrogram.webp ONLY reflect the part starting at beat 113.

Please analyze the attached files and provide concrete mixing advice.

---

🛠️ Development & Contributing

Local Setup (from source)

git clone https://github.com/peterkolbe/ableton-for-ai.git

cd ableton-for-ai uv sync

Run the MCP server from source:

{

"mcpServers": { "ableton-for-ai": { "command": "uv", "args": [ "--directory", "/absolute/path/to/ableton-for-ai", "run", "mcp_server_ableton.py" ], "env": { "STEMS_SOURCE_DIR": "/path/to/your/stems" } } } }

Run CLI commands from source:

uv run ableton_client.py analyze_stems

uv run ableton_client.py extract_ableton_project_data uv run ableton_client.py analyze_stems_and_extract_ableton_project_data

MCP Inspector (Debugging)

The MCP Inspector is useful for testing and debugging the MCP server during development.

Start it directly via terminal:

npx -y @modelcontextprotocol/inspector uv '--directory /absolute/path/to/ableton-for-ai run mcp_server_ableton.py'

Or start the Inspector without presets and configure it manually:

npx -y @modelcontextprotocol/inspector

Then enter the following in the Inspector UI:

• Command: uv
• Arguments: --directory /absolute/path/to/ableton-for-ai run mcp_server_ableton.py

Linting & Formatting

uv run ruff check .

uv run ruff format .

Publishing to MCP Registry

To publish a new version to the official MCP Registry:

brew install mcp-publisher

mcp-publisher login github mcp-publisher publish

Verify the publication:

curl "https://registry.modelcontextprotocol.io/v0/servers?search=io.github.peterkolbe/ableton-for-ai"

Make sure to update the version in both server.json and pyproject.toml before publishing.

---

⚠️ Known Limitations & Troubleshooting

Automation Trac