How do I give Claude, ChatGPT or Cursor access to podcast transcripts?

The simplest way to give an AI assistant access to podcast transcripts is an MCP connector. Add https://mcp.audioscrape.com as a custom connector in Claude, ChatGPT or Cursor, sign in with a free Audioscrape account, and the assistant can search more than 18 million transcribed episodes, pull full transcripts with speaker labels and timestamps, and cite the exact moment. No files to download, no pipeline to build.

Why an MCP server rather than pasting transcripts

You can paste a transcript into a chat, but a two-hour episode is tens of thousands of words, most of it irrelevant to your question, and you first have to obtain the transcript. Over MCP the assistant calls a search tool, gets back only the matching segments with speaker, timestamp and source link, and fetches a full transcript only when it needs one. That keeps the context small and every quote traceable.

How to do it with Audioscrape

Claude (web or desktop)

1. Requires a Claude Pro, Max, Team or Enterprise plan; custom connectors are not on Claude’s free tier.
2. Settings, then Connectors, then Add custom connector. Paste https://mcp.audioscrape.com.
3. Audioscrape appears in the Search and tools menu. Click Connect. Sign in with Google or Microsoft; if you are already signed in on audioscrape.com you land directly on the authorize screen.
4. Ask: “Find recent podcast discussions about the Fed’s rate decision, with timestamps.” Guide: Claude Desktop. Claude Code users can add it with a one-line command: Claude Code.

ChatGPT

1. Requires Plus, Pro, Team, Enterprise or Edu. Settings, Connectors, Advanced, and turn on Developer Mode (Team and Enterprise admins enable it under Workspace Settings first).
2. Settings, Connectors, Create. Enter https://mcp.audioscrape.com as the MCP server endpoint, give it the name Audioscrape and a short description; ChatGPT uses these to decide when to call it.
3. Authorize with your Audioscrape account.
4. In a chat, click the plus icon, More, Developer mode, and enable Audioscrape for that conversation. Guide: ChatGPT.

Cursor

Works on every Cursor plan including free. Settings, MCP and Integrations, New MCP Server, then paste:

{
  "mcpServers": {
    "audioscrape": {
      "url": "https://mcp.audioscrape.com"
    }
  }
}

The first time you call a tool, a browser window opens for sign-in. Guide: Cursor.

Any other MCP client can point at the same URL. Authentication is OAuth 2.1 with PKCE and dynamic client registration, handled automatically by spec-compliant clients. For headless agents, create an API key from your workspace’s API Keys tab and send it as a Bearer token to https://mcp.audioscrape.com/; the MCP overview has a curl quickstart.

What the assistant can do once connected

The server exposes tools for the full record. The ones you will see most:

• search_audio: keyword or semantic search over every transcript, public plus your workspace.
• get_episode_overview: chapters, speakers and top entities for one episode, the cheap way to summarize.
• get_transcript: the full transcript with speaker labels, paged for long episodes.
• search_speakers and get_speaker: find a person and list their appearances.
• search_podcasts, list_podcast_episodes, list_recent_transcripts: navigate shows.
• search_entities, get_entity_graph, get_trending, get_charts: the knowledge graph and what is being discussed now.
• transcribe_audio, get_transcription_status, list_my_uploads: bring your own recordings.
• create_share_link: a shareable URL for a specific moment.

Every result carries the speaker, a timestamp and a source link, so the assistant can quote with a citation you can check.

Limits and honest caveats

• The free Audioscrape account includes 100 searches per month and 30 transcription minutes; MCP and REST share one quota. Semantic search is a Pro and Enterprise feature; text search is on every plan. There are no free trials of paid plans.
• Search returns segments, not whole episodes. Assistants should call get_transcript when they need everything.
• The index covers spoken audio only. It is not a web search engine.
• Not every podcast is indexed, and new episodes are transcribed in batch, so very recent releases may lag.
• Assistants sometimes do not pick the tool on their own. Saying “use Audioscrape to…” fixes that.
• Speaker names in the public corpus are attributed automatically and can be reported if wrong.

Doing it without a connector

If you only need one episode, get the transcript from the show’s own site if it publishes one, or transcribe a file you have the rights to, and paste the relevant part into the chat. For anything across multiple shows or over time, a searchable index is the practical route.

Plans and limits: /pricing. Facts for AI agents: /docs/for-agents.