
通过 MCP 通过 AI 助手控制 Ableton Live。添加轨道、设备、剪辑、自动化和生成 MIDI,按音符概率计算。
概览
通过像Claude这样的AI助手控制Ableton Live。你用通俗易懂的语言描述你想要的内容,它会在你的Live集合中构建:轨道、乐器、效果、MIDI剪辑、自动化和混音。超过150个命令涵盖了大部分Live Object Model。 你描述,它积累,你倾听并评判。助理没有耳朵,所以它负责机械工作,而你保持味道。 提出那些你绝不会亲手做的繁琐事情: 助手通过模型上下文协议(MCP)与 Ableton 通信,MCP 是一个开放标准,用于让 AI 工具访问外部系统。该项目是一个 MCP 服务器,将 Ableton 的 Live Object Model 作为助手可调用的工具集合展示。 MCP服务器作为独立进程运行在您的机器上。您的AI客户端(Claude Code、Claude Desktop、Cursor)启动它并调用其工具。 远程脚本运行在Ableton内部。
概述(AI 翻译)
通过像Claude这样的AI助手控制Ableton Live。你用通俗易懂的语言描述你想要的内容,它会在你的Live集合中构建:轨道、乐器、效果、MIDI剪辑、自动化和混音。超过150个命令涵盖了大部分Live Object Model。
你描述,它积累,你倾听并评判。助理没有耳朵,所以它负责机械工作,而你保持味道。
目录
- 你能用它做什么
- 工作原理
- 安装
- 开始
- 技能
- 使用局部模型代替Claude
- 编辑远程脚本
- 安全
- 测试
- 建筑学
- 制作人员
- 许可证
你能用它做什么
要求一个声音并让它制作:
提出那些你绝不会亲手做的繁琐事情:
要求一整件风格的作品:
请它修正你刚才听到的内容:
它是如何运作的
助手通过模型上下文协议(MCP)与 Ableton 通信,MCP 是一个开放标准,用于让 AI 工具访问外部系统。该项目是一个 MCP 服务器,将 Ableton 的 Live Object Model 作为助手可调用的工具集合展示。
共有两部分:
MCP服务器作为独立进程运行在您的机器上。
本页由 AI 翻译,技术细节请以原文为准。
README
以下为原文 README 节选,完整内容请查看原文。
Ableton AI
Control Ableton Live from an AI assistant like Claude. You describe what you want in plain language, and it builds it in your Live set: tracks, instruments, effects, MIDI clips, automation, and mixing. Over 150 commands covering most of the Live Object Model.
You describe, it builds, you listen and judge. The assistant has no ears, so it handles the mechanical work while you keep the taste.
Contents
- What you can do with it
- How it works
- Installation
- Getting started
- The skill
- Using a local model instead of Claude
- Editing the remote script
- Safety
- Testing
- Architecture
- Credits
- License
What you can do with it
Ask for a sound and get it built:
Build a dark ambient pad on a new track. Slow attack, long release,
a filter behind it with a slow LFO on the cutoff. Play a minor chord.
Ask for the tedious things you’d never do by hand:
Write six clips at 3, 5, 7, 11, 13 and 17 bar lengths. They're coprime,
so the loop won't repeat for over a million bars.
Draw a twelve minute filter sweep with 200 automation points, then put
per-note probability on the hats so they flicker instead of repeating.
Ask for a whole piece in a style:
Lay down a Steve Reich style phasing piece. Two piano tracks play the same
short cell; make one clip a bar longer than the other so they drift out of
phase and slowly pull back into sync. Add a marimba and a vibraphone doubling
fragments of the cell at their own loop lengths, so the parts interlock and
never quite repeat.
Ask it to fix what you just heard:
The bass is too quiet and the reverb is too wet. Pull them back.
How it works
The assistant talks to Ableton through the Model Context Protocol (MCP), an open standard for giving AI tools access to external systems. This project is an MCP server that exposes Ableton’s Live Object Model as a set of tools the assistant can call.
There are two pieces:
The MCP server runs on your machine as its own process. Your AI client (Claude Code, Claude Desktop, Cursor) launches it and calls its tools.
A remote script runs inside Ableton itself. Ableton’s Control Surface system lets you run Python inside Live, and that script opens a local socket the MCP server talks to. This is the only supported way into Live’s API, so it has to be Python and it has to be enabled in Ableton’s settings.
When you ask for something, the assistant calls a tool, the server sends a command over the socket, the remote script runs it against Live’s API, and the result comes back. You see it happen in Ableton in real time.
Installation
You need Ableton Live 11 or newer, and uv.
安装
uvx ableton-ai配置
{
"mcpServers": {
"ableton": {
"command": "uvx",
"args": ["ableton-ai"]
}
}
}