I’ve been working on a way to connect Claude (and other Model Context Protocol clients like ChatGPT) to a large index of public and user-uploaded audio.

The result is a Remote MCP server for Audioscrape that lets an AI assistant query and return relevant segments from:

Popular podcasts

Court hearings

Earnings calls

Any audio you upload yourself

Example query:

"What is Joe Rogan saying about Germany on his podcasts?"

The MCP server handles the transcription index and exposes a search_audio_content tool. Claude then calls it and surfaces exact matching moments from the audio.

Tech notes:

Server: Remote MCP spec implementation

Search: Full-text index over transcripts

Client: Works with Claude Desktop, Mobile, Web MCP support

Data: Combination of public audio sources + user uploads

Docs & setup:

Claude setup: https://www.audioscrape.com/docs/claude-setup

MCP server details: https://www.audioscrape.com/docs/mcp

I’m curious how others might extend MCP for searching non-text data sources - especially if you’ve tried integrating AI assistants with large datasets outside the usual APIs.