fmod-mcp
An MCP server that enables AI coding agents to control FMOD Studio for audio import, event creation, and bank building via TCP scripting.
README
fmod-mcp
An MCP server that lets an AI coding agent drive FMOD Studio: import audio, create events, wire instruments, place effects on buses, set up sends/returns, manage parameters, and build banks. It does this by speaking to Studio's built-in JavaScript scripting terminal over TCP (port 3663).
v0.2: 35 tools, 143 tests passing (106 unit + 37 live, plus 1 gated persist test). See CLAUDE.md for the tool catalog, docs/effects.md for the effect-parameter reference, docs/troubleshooting.md when something breaks, and examples/ for worked walkthroughs.
Built for use with Claude Code, but works with any MCP-compatible client.
Requirements
- FMOD Studio 2.02+ (the scripting terminal has been stable across recent versions)
- Python 3.11+
- A running FMOD Studio instance with the target project open
Install
git clone https://github.com/jmperez127/fmod-mcp.git
cd fmod-mcp
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
Register with Claude Code
Run this from inside the cloned fmod-mcp directory (so $(pwd) resolves correctly):
claude mcp add fmod -s user -- "$(pwd)/.venv/bin/python" -m fmod_mcp
The -- ends claude mcp add's flag parsing so -m fmod_mcp is treated as
arguments to Python. -s user registers the server for every Claude Code
session on your machine; drop it for a project-local registration.
Alternatively, if you prefer a globally-callable shim, install via pipx:
pipx install .
claude mcp add fmod -s user fmod-mcp
Verify:
claude mcp list
Configure host/port (optional)
Defaults to 127.0.0.1:3663. Override with environment variables:
export FMOD_MCP_HOST=127.0.0.1
export FMOD_MCP_PORT=3663
Enable TCP scripting in Studio
FMOD Studio's scripting terminal listens on TCP/3663 by default. If it's disabled:
- Open FMOD Studio
Edit→Preferences(orFMOD Studio→Preferenceson macOS)- Look for
Scripting/Terminalsettings - Enable TCP listener, note the port
Quick smoke test
With FMOD Studio running and a project open:
> ping
{ ok: true, version: { productVersion: 2, majorVersion: 2, minorVersion: 20, ... } }
> list_banks
[ { path: "bank:/Master", guid: "...", name: "Master" }, ... ]
Worked examples
The examples/ directory has prompt-style walkthroughs you can paste into a Claude Code session:
| File | Scenario |
|---|---|
| add_one_sfx.md | Minimum pipeline: import → event → sound → save → build |
| batch_import_sfx.md | Loop the above over a folder of wavs |
| add_reverb_to_bus.md | Reverb return + global send (with parameter reference) |
| sidechain_compression.md | Music-ducks-to-VO routing |
| parameter_driven_volume.md | Local game parameter for an event |
| cleanup_unused_audio.md | Find and delete orphan audio in the bin |
Files it writes
~/.cache/fmod-mcp/commands.log: every JS snippet sent to Studio, timestamped by request ID. Audit trail + replay source.
Testing
Unit tests run without Studio (mock-based):
pytest -q
# 106 passed, 38 skipped (live tests)
Live tests require FMOD Studio running with any project open:
FMOD_MCP_LIVE=1 pytest -q
# 143 passed, 1 skipped (persist-gated)
The persist-gated save+build verification additionally writes to disk; opt in explicitly:
FMOD_MCP_LIVE=1 FMOD_MCP_LIVE_PERSIST=1 pytest tests/live/test_save_build_live.py
Live tests use a per-test scratch folder under event:/__mcp_test_scratch__/<hex>/ and clean up via studio.project.deleteObject on tear-down. A session-scoped sweep also runs at start AND end so an interrupted run doesn't leave orphans.
When things break
See docs/troubleshooting.md for the common failure modes and their fixes. The short version:
- Check
~/.cache/fmod-mcp/commands.logfor the exact JS that went out. - Paste it into Studio's own Scripting window (
Window→Scriptingin the Studio GUI) and iterate. - Once working, patch the corresponding JS template in
fmod_mcp/tools/*.py.
For anything not covered by a named tool, use run_js:
> run_js return studio.project.model.Event.findInstances().length;
42
Layout
fmod_mcp/
├── studio_client.py # TCP/3663 client. Single persistent connection, sentinel-framed IIFE protocol
├── server.py # FastMCP tool registration, stdio transport
├── __main__.py # python -m fmod_mcp
└── tools/
├── discovery.py # ping, list_banks, list_events, list_buses, get_event
├── audio.py # import_audio, delete_audio, move_audio
├── events.py # create_event, add_single_sound, add_multi_sound,
│ # set_event_property, assign_to_bank, assign_to_bus,
│ # delete_event, delete_folder, rename_event, move_event,
│ # add_local_parameter, add_global_parameter
├── effects.py # list_effect_types, add_effect, list_effects, get_effect,
│ # set_effect_param, remove_effect, bypass_effect
├── routing.py # add_return, remove_return, list_returns,
│ # add_send, list_sends, remove_send, set_send_level
├── project.py # save_project, build_banks
└── escape.py # run_js
tests/
├── (mock-based unit tests)
└── live/ # opt-in via FMOD_MCP_LIVE=1; auto-skip when port 3663 isn't reachable
License
MIT © 2026 jmperez127.
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。