fcp-midi
Enables LLMs to compose music by describing musical intent using high-level operations like notes, chords, dynamics, and tempo, then renders the composition into standard MIDI files.
README
fcp-midi
MCP server for semantic MIDI composition.
What It Does
fcp-midi lets LLMs compose music by describing musical intent -- notes, chords, dynamics, tempo changes -- and renders it into standard MIDI files. Instead of manipulating raw bytes or MIDI events, the LLM works with operations like note Bass E2 at:1.1 dur:quarter vel:90 and crescendo @track:Lead vel:45-75 @range:3.1-6.4. Built on the FCP framework, powered by pretty-midi for serialization.
Quick Example
midi_session('new "Voltage Drop" tempo:140 key:E-minor')
midi([
'note Drums kick at:1.1 dur:eighth vel:100',
'note Bass E2 at:1.1 dur:quarter vel:90',
'chord Pad Em at:1.1 dur:whole vel:70',
'crescendo @track:Lead vel:45-75 @range:3.1-6.4',
'tempo 174 at:11.1',
])
midi_session('save as:./voltage_drop.mid')
Available MCP Tools
| Tool | Purpose |
|---|---|
midi(ops) |
Batch mutations -- notes, chords, tracks, tempo, dynamics, copy/transpose |
midi_query(q) |
Inspect the composition -- map, tracks, events, piano-roll, instruments, find |
midi_session(action) |
Lifecycle -- new, open, save, checkpoint, undo, redo |
midi_help() |
Full reference card |
Hero Examples
Plumber's Journey -- Classic game theme faithfully recreated: 4 tracks, 288 notes, 16 seconds, 180 BPM.
Voltage Drop -- A drum-and-bass track with tempo acceleration (140 -> 155 -> 174 BPM), 5 tracks, breakbeats, sub-bass, arpeggios, and a signature DROP section. Also used in the FCP vs raw Python agent battle -- FCP produced 1,674 notes across 12 tracks in 87 seconds, compared to 694 notes from ~689 lines of hand-written Python.
See docs/examples/ for MIDI files and the full writeup.
Installation
Requires Python >= 3.11.
pip install fcp-midi
MCP Client Configuration
{
"mcpServers": {
"midi": {
"command": "uv",
"args": ["run", "python", "-m", "fcp_midi"]
}
}
}
Architecture
3-layer architecture:
MCP Server (Intent Layer)
Parses op strings, dispatches to verb handlers
|
Semantic Model
Tracks, notes, chords, dynamics, tempo maps, markers
In-memory composition graph with event sourcing
|
Serialization (pretty-midi)
Semantic model -> MIDI binary output
Key features:
- Instrument library -- GM instruments by name (
acoustic-grand-piano,synth-bass-1,standard-kit) - Soundfont support -- Load custom instrument banks
- Beat addressing --
at:1.1(bar 1, beat 1),at:3.2.240(bar 3, beat 2, tick 240) - Duration vocabulary --
whole,half,quarter,eighth,sixteenth,triplet - Dynamics --
crescendo,diminuendoacross ranges - Copy/transpose -- Duplicate and shift musical phrases
Development
uv sync
uv run pytest # 658 tests
uv run pytest -m "not slow" # skip stress tests
uv run ruff check # linting
uv run pyright # type checking
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。