Skip to content

MCP Documentation

MCP AI Song Detector Server

The AI Song Detector MCP server lets AI applications — from autonomous agents to chatbots and other assistants — check whether an audio file contains AI-generated music directly from a conversation, without writing any API integration code.

It wraps the same detection capability described in the API Documentation: submit a song, get back whether it is AI-generated, a confidence score, and the predicted generation platform (e.g. Suno, Udio). The model produces one prediction for the entire file.

The MCP server is available at:

https://mcp.pex.com/mcp

MCP Tools

The server exposes two tools:

  • detect_ai_song_from_url — Detects whether audio retrieved from a URL is AI-generated.

    • url (string) — publicly accessible URL of the audio file.

  • detect_ai_song_from_base64 — Detects whether base64-encoded audio is AI-generated.

    • file_contents (string) — base64-encoded audio data.
    • ⚠️ Use only for small files. The base64-encoded audio is passed through the model's context window, so anything but a small clip can quickly exhaust the available tokens and cause the request to fail. For larger files, prefer detect_ai_song_from_url.

Both tools return a JSON object with the same fields as the API — is_ai, ai_score, predicted_model, and more. See Response Format for details.

MCP Authentication

The MCP server uses OAuth 2.0. When you add the connector to your AI agent, you will be asked for a Client ID and Client Secret, and the agent will show an authorization consent page.

To obtain a Client ID and Client Secret, please contact our sales team.

Authorization has been tested with the following clients:

  • Claude
  • Grok — set Token Auth Method to client_secret_post or client_secret_basic
  • Gemini CLI
  • ChatGPT

Support

If you encounter problems, please contact your Vobile representative. See the API Documentation for the details to include in your report.