MCP-Audio Plugin

`mcp-audio` is an AIO-2030 compliant MCP plugin that performs voice-to-text transcription using the Audio speech recognition API.

It exposes the `identify_voice` method via both `multipart/form-data` and `base64` formats, supports the AIO `tools.call` protocol, and returns JSON-RPC structured outputs.

Features

• Fully AIO-compliant MCP plugin (`/tools.call`, `/help`)
• Converts `.wav`/`.mp3` audio files to transcripts using SiliconFlow
• API key managed securely via `.env` file
• Docker-compatible and minimal dependencies
• Registration-ready for AIO endpoint registry

Setup (Local)

1. Clone and Install

git clone git@github.com:AIO-2030/mcp-audio.git
cd mcp-audio
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

2. Add .env file

cp .env.example .env

Set your audio URL and API key:

AUDIO_URL=https--xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

3. Run the MCP server

python src/mcp_server.py

4. Docker

4.1 Build and Run

docker build -t mcp-audio .
docker run --env-file .env -p 8080:8080 mcp-audio

API Overview

POST /api/v1/mcp/voice_model

Upload audio file directly. Response:

{
  "transcript": "hello world",
  "confidence": 0.91,
  "audio_hash": "a1b2c3..."
}

POST /api/v1/mcp/tools.call (AIO Protocol)

JSON-RPC format with base64-encoded audio. Response:

{
  "method": "tools.call",
  "params": {
    "method": "identify_voice",
    "inputs": [
      {
        "type": "audio",
        "value": ""
      }
    ]
  }
}

GET /api/v1/mcp/help

Auto-serves contents of mcp_audio_registration.json. Used by Queen AI for MCP discovery and service indexing.

Testing Tools

Base64 Voice Test

python test/test_audio_base64.py

Health Check

python health_check.py

MCP Registration (to AIO Endpoint Canister)

./register_mcp.sh

Requires jq, dfx, and a running endpoint_registry canister.