svs-mcp
An MCP server for Synthesizer V Studio 2 Pro on macOS that enables AI-driven composition and lyric input, including melody creation, lyric flowing, and phoneme tuning.
README
svs-mcp
An MCP (Model Context Protocol) server for Synthesizer V Studio 2 Pro, focused on AI-driven input of lyrics and notes: compose melodies, flow lyrics onto them, and fine-tune pronunciation — all from an AI client such as Claude.
- macOS + SV Studio 2 Pro only. Local build only (not published to npm).
- Design notes and background live in docs/plan.md.
How it works
SV's scripting API has no network access, so a small Lua script runs inside SV Studio and exchanges JSON files with the MCP server:
AI client (Claude, etc.)
│ MCP (stdio)
▼
svs-mcp server (Node.js)
│ request.json / response.json in ~/.svs-mcp/bridge/
▼
SVSMCPBridge.lua (resident script inside SV Studio, polls every 100ms)
Requests carry unique IDs (stale responses are discarded) and all file writes are atomic (temp file + rename).
Setup
Requirements: Node.js 20+, Synthesizer V Studio 2 Pro on macOS.
npm install
npm run build
npm run install-sv # copies the bridge scripts into SV's scripts folder
Then, in SV Studio:
- Run Scripts > Rescan (or restart SV Studio).
- Run Scripts > SVS MCP > "SVS MCP Bridge: Start". A dialog confirms the bridge is running; it keeps running in the background.
Finally, register the server with your MCP client:
{
"mcpServers": {
"svs-mcp": {
"command": "node",
"args": ["/path/to/svs-mcp/dist/index.js"]
}
}
}
To verify the setup without an MCP client, run npm run ping — it prints SV's
host info when the bridge answers, or setup guidance when it doesn't (exit 1),
which also makes the whole setup scriptable for an AI agent: only the two
Scripts-menu clicks above need a human.
To stop the bridge, run "SVS MCP Bridge: Stop" from the Scripts menu.
To remove everything, npm run uninstall-sv.
Tools
| Tool | Purpose |
|---|---|
ping |
Check the bridge is alive; returns SV host info |
get_project_info |
Project overview: tempo, time signatures, tracks, groups, mixer state |
get_notes |
List notes (lyrics, pitch, position) of a track/group, filterable by measure |
add_track |
Add a track |
create_group |
Add a note group to a track |
insert_notes |
Insert notes: lyrics + pitch ("C4") + musical position (measure/beat) + note value ("1/8") |
update_notes |
Edit lyrics/pitch/timing of existing notes by index |
delete_notes |
Delete notes by index |
set_lyrics |
Flow lyrics (one syllable per note) onto an existing melody |
get_phrases |
Analyze a melody into phrases/sections with repetition (shape) labels |
set_time_signature |
Add/update/remove time signature marks (align SV's grid with the music) |
open_lyrics_editor |
Browser page for the user to adjust syllable boundaries before applying |
get_lyrics_editor_result |
Fetch the user-edited syllable plan from the editor |
get_phonemes |
Inspect the phonemes SV will actually sing |
set_phonemes |
Override phonemes per note ("k a"); empty string resets |
set_language |
Per-note language override for mixed-language lyrics |
Positions and durations use musical notation (1-based measures/beats, note values
like 1/8 or dotted 1/8.); the server converts to SV's internal blicks.
Japanese lyrics take one kana per note; - extends the previous vowel.
For putting lyrics on an imported MIDI melody, see docs/lyrics-workflow.md.
Important: groups and singers
In SV Studio 2 the singer is attached per note group, and notes placed in a track's main group are not synthesized (they appear in the piano roll but stay silent). Therefore:
insert_notestargets the first non-main group by default and auto-creates a group when the track has none.- New tracks/groups inherit SV's default singer. There is no scripting API for singer selection, so if a group has no singer, pick one in the SV UI (select a note of the group, then use the Voice panel).
Troubleshooting
- Tool calls time out — the bridge is probably not running. Run "SVS MCP Bridge: Start" in SV Studio. Only one MCP server instance should talk to one SV instance.
- Notes appear but don't sound — check that the notes are in a non-main group
(
get_project_infoshows the groups) and that the group has a singer assigned in the Voice panel. Also check track/group mute states, whichget_project_inforeports. - Wrong pronunciation — inspect with
get_phonemes, then correct withset_phonemesorset_language.
Development
npm test # unit + protocol tests + e2e against the real Lua bridge (needs `lua` on PATH)
npm run build # tsc
test/sv-stub.lua emulates the SV host so the actual bridge script can be tested
without SV Studio. After changing sv-scripts/, re-run npm run install-sv and
restart the bridge from SV's Scripts menu.
License
MIT. Bundles rxi/json.lua (MIT) inside the bridge script.
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。