MCP Architecture

The Model Context Protocol (MCP) server is the primary interface for AI assistants (like Claude) to interact with your Audio Knowledge Base.

🛠️ Available Tools

1. search_audio

Semantically search your transcription knowledge base.

• Arguments: query (string), limit (int).
• Returns: List of matching segments with timestamps and sources.

2. ingest_url

Submit a public audio URL for background transcription and indexing.

• Arguments: url (string).
• Returns: Confirmation and Task ID.

3. get_task_status

Monitor the progress of a transcription job.

• Arguments: audio_id (uuid).
• Returns: Current status (queued, processing, completed).

📚 Resources

The server exposes transcripts as MCP resources:

• transcript://{audio_id}: Access the full raw transcript of a processed audio file.

🛡️ Security & Auth

The MCP server uses standard X-API-Key authentication. For local use with Claude Desktop, the key is passed via environment variables in the config file.

[!TIP] Always use a dedicated API key for your MCP integration to track usage independently in the dashboard.