MCP 服务器

AI Sound MCP Server

Enables AI assistants to programmatically edit, analyze, and export audio projects through MCP tools, including multi-track editing, effects, transcription, and semantic search.

README

An AI-native audio editor that works with any OpenAI-compatible LLM.

What is AI Sound?

AI Sound is an AI-native audio editor designed as a modern replacement for desktop tools like Audacity. Instead of bolting AI onto an existing app, AI Sound is built from the ground up with LLM integration at its core — enabling conversational audio editing, automatic transcription, speaker diarization, and semantic search across your audio content.

It works with any OpenAI-compatible API — run it fully local with Ollama, or connect to OpenAI, Groq, or any other compatible provider. No vendor lock-in, no API keys required for local use.

AI Sound also exposes a full MCP (Model Context Protocol) server, letting AI assistants like Claude Desktop directly edit, analyze, and export your audio projects.

Key Features

Multi-track editing — import, arrange, and mix multiple audio tracks
AI-powered transcription — speech-to-text with word-level timestamps
Speaker diarization — automatically identify and split by speaker
Semantic search — find content by meaning, not just keywords
Audio effects — normalize, compress, EQ, reverb, noise reduction, fade, pitch shift, speed
Non-destructive editing — full undo/redo history
Export — export individual tracks or full project mixes
MCP integration — expose all editing tools to AI assistants

Quickstart

Prerequisites

Bun — JavaScript/TypeScript runtime
FFmpeg — audio processing (brew install ffmpeg on macOS)
An OpenAI-compatible LLM — Ollama for local, or any cloud provider

Install & Run

git clone https://github.com/your-username/ai-sound.git
cd ai-sound
bun install
bun run dev

Open http://localhost:5175 in your browser.

On first launch, configure your LLM provider in Settings (gear icon). The default is Ollama at localhost:11434.

LLM Configuration

AI Sound works with any OpenAI-compatible API. Configure your provider in-app via Settings — no .env files needed.

Provider	Base URL	Model Example
Ollama (local)	`http://localhost:11434/v1`	`llama3.2`
OpenAI	`https://api.openai.com/v1`	`gpt-4o`
Anthropic (via proxy)	provider-specific	`claude-sonnet-4-20250514`
Groq	`https://api.groq.com/openai/v1`	`llama-3.3-70b`

For cloud providers, enter your API key in the Settings panel. For Ollama, no API key is needed.

MCP Server

AI Sound includes a built-in Model Context Protocol server, allowing AI assistants like Claude Desktop to interact with your audio projects programmatically.

Claude Desktop Configuration

Add the following to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "ai-sound": {
      "command": "bun",
      "args": ["run", "/absolute/path/to/ai-sound/server/lib/mcp/server.ts"]
    }
  }
}

Replace /absolute/path/to/ai-sound with the actual path to your installation.

Available Tools

Project Management

list_projects — List all projects
set_active_project — Set the active project for subsequent operations
get_project_status — Get all tracks, durations, regions, and transcriptions
get_track_info — Get detailed info about a specific track

Audio Effects & Processing

normalize_audio — Normalize audio levels
adjust_volume — Adjust volume by relative dB amount
trim_audio — Trim to a specific time range
remove_silence — Detect and remove silent sections
apply_fade — Apply fade in/out
apply_effect — Apply effects: noise reduction, compressor, EQ, reverb, speed, pitch shift

Segment Operations

remove_segments — Remove multiple time ranges from a track
replace_audio_segment — Replace a time range with silence or a beep

Track Operations

rename_track — Rename a track
delete_track — Delete a track and its audio
merge_tracks — Merge multiple tracks into one
duplicate_track — Duplicate a track
export_audio — Export a track or full project mix

Transcription & Search

transcribe_track — Transcribe audio using speech-to-text
split_by_speaker — Split a track by speaker
rename_speaker — Rename a speaker label
search_transcription — Search transcription text by pattern
search_transcript_semantic — Semantic search across transcriptions
copy_transcriptions — Copy transcription data between tracks

Tech Stack

Layer	Technology
Runtime	Bun
Server	Hono
Database	SQLite via Drizzle ORM
Frontend	React 19 + Tailwind CSS 4
Audio	wavesurfer.js + FFmpeg
AI Integration	OpenAI-compatible API + MCP SDK