Keyboardia MCP
Enables agents to co-edit live music sessions, create, remix, publish, export MIDI, and analyze musical structure in the Keyboardia collaborative step sequencer.
README
Keyboardia
A multiplayer step sequencer with polyrhythmic patterns, built for real-time collaboration.
Features
- Real-time Multiplayer - Up to 10 players can jam together on the same session
- Polyrhythmic Patterns - Each track can have 3-128 step counts (26 options including triplet-friendly values)
- 70 Sound Generators - 32 Web Audio synths, 11 Tone.js FM/AM synths, 27 sampled instruments
- 27 Sampled Instruments - Grand piano, 808 kit, acoustic drums, Hammond organ, kalimba, steel drums, strings, guitars, sax, and more
- Effects Chain - Reverb, delay, chorus, and distortion with limiter (full multiplayer sync)
- Parameter Locks - Per-step pitch, volume, and tied notes automation
- Chromatic Grid - Two view modes: "Events" (key intervals + used pitches) and "All" (49 pitches from -24 to +24)
- Scale Lock - Constrain chromatic grid to selected musical scale with out-of-scale warnings
- Scale Sidebar - Visualize scale notes with root/fifth emphasis and active usage highlighting
- Per-track Swing - Global and per-track swing settings for groove control
- Session Sharing - Share links, remix others' work, publish immutable sessions
- QR Code Sharing - Mobile-friendly session sharing
- Agent Rhythm Editing (Experimental) - Co-edit, create, remix, publish, export, and analyze sessions through stateless MCP
Use with an agent
Configure your MCP client with:
https://keyboardia.dev/mcp
To co-edit music you already have, open a session and give the agent the UUID
from its https://keyboardia.dev/s/{session_id} URL. Agents can read the
current rhythm, add a track, assign specific steps, and change tempo. Their
edits use the same live session as connected browsers, and published sessions
stay read-only.
Agents can also start from nothing: create a new session and hand back its link, remix a published session into an editable copy without touching the original, publish the current result as an immutable snapshot when you ask, and export the session as a MIDI file for a DAW.
They can explain music too — ask what key a session is in, how its rhythms sit against each other, or what chord a moment forms, and the answer comes from the same music-theory module the Key Assistant uses, with its uncertainty stated rather than hidden.
See the stateless MCP rhythm-slice specification for the exact tool contract and current limitations. Directory maintainers can use the canonical MCP listing record, and hosted data handling is described in the MCP privacy notice.
Tech Stack
- Frontend: React 19, TypeScript, Vite
- Audio: Web Audio API, Tone.js
- Backend: Cloudflare Workers, Durable Objects, KV Storage
- Real-time: WebSockets with Hibernation API
Getting Started
cd app
# Install dependencies
npm install
# Start development server
npm run dev
# Run tests
npm run test:all
# Build for production
npm run build
Development
Debug Mode
Add ?debug=1 to the URL to enable debug features:
- Event tracing
- Audio state debugging
- Persistent log storage
- Bug pattern detection
See docs/DEVELOPMENT-TOOLS.md for comprehensive debugging documentation.
Project Structure
app/
├── src/
│ ├── audio/ # Audio engine, synths, effects, scheduling
│ ├── components/ # React UI components
│ ├── hooks/ # React hooks (useSession, useMultiplayer, etc.)
│ ├── state/ # State management (grid reducer)
│ ├── sync/ # Multiplayer synchronization
│ ├── worker/ # Cloudflare Worker (Durable Objects, API routes)
│ ├── utils/ # Logging, debugging, utilities
│ └── debug/ # Debug overlay and context
├── e2e/ # End-to-end tests (Playwright)
├── test/ # Integration tests
└── scripts/ # Development and debugging scripts
Key Files
| File | Description |
|---|---|
app/src/audio/engine.ts |
Main audio engine - coordinates all audio subsystems |
app/src/audio/scheduler.ts |
Drift-free lookahead scheduling (25ms timer, 100ms lookahead) |
app/src/audio/synth.ts |
16-voice polyphonic synthesizer with voice stealing |
app/src/audio/toneSynths.ts |
Tone.js synth manager (FM, AM, Membrane, etc.) |
app/src/audio/toneEffects.ts |
Effects chain (reverb, delay, chorus, distortion) |
app/src/sync/multiplayer.ts |
WebSocket client for real-time sync |
app/src/worker/live-session.ts |
Durable Object for session state |
Testing
cd app
npm run test:unit # Unit tests (vitest)
npm run test:integration # Integration tests (Cloudflare Workers)
npm run test:all # All tests
npm run analyze:bugs # Static bug pattern analysis
Architecture
Audio Signal Chain
Source (Oscillator/Sample)
→ Track Gain (per-track volume)
→ Master Gain
→ Effects Chain (Tone.js: reverb → delay → chorus → distortion)
→ Limiter
→ Compressor
→ Destination
Synth Engines
- SynthEngine (
synth.ts) - Native Web Audio oscillators, 40+ presets - ToneSynthManager (
toneSynths.ts) - Tone.js FM/AM/Membrane synths - AdvancedSynthEngine (
advancedSynth.ts) - Dual-oscillator with filter envelope and LFO - SampledInstrument (
sampled-instrument.ts) - Sample-based playback (piano)
Multiplayer Architecture
Client A ←→ Durable Object ←→ Client B
↓
DO Storage (immediate)
↓
KV Storage (on disconnect)
- Each session is a single Durable Object instance
- WebSocket connections use Hibernation API for cost efficiency
- State changes broadcast to all connected clients
- Hybrid persistence: Mutations saved immediately to DO storage, KV updated on disconnect
Documentation
- ROADMAP.md - Implementation phases and status
- SYNTHESIS-ENGINE.md - Audio architecture spec
- SHARING-AND-PUBLISHING.md - Session persistence spec
- DEVELOPMENT-TOOLS.md - Debug tools reference
- UI-PHILOSOPHY.md - Design principles
- LESSONS-LEARNED.md - Debugging war stories
License
Private - All rights reserved
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。