Is there an API for podcast transcripts?

Yes. Audioscrape provides a REST API at https://www.audioscrape.com/api/ and an MCP server at https://mcp.audioscrape.com that give programmatic access to speaker-attributed, timestamped transcripts of more than 18 million podcast episodes, plus transcripts of audio you upload. You authenticate with a Bearer API key, search with POST /api/search, and every result carries the speaker, the start and end time, and a link to the source audio. The full OpenAPI spec is at /api/openapi.json.

What kind of API you actually want

“Podcast transcript API” covers three different needs, and it helps to be clear which one you have:

• Transcribe one file I supply. That is a speech-to-text API. Audioscrape can do it as part of an upload, but it is not a per-request transcription endpoint and it is not real-time.
• Read the transcript of a public episode. Audioscrape has already transcribed it; you fetch it.
• Search across everything said on podcasts. This is the core of the product: a search index over the spoken words, with filters for podcast, speaker and date.

If you need streaming captions for live audio, use a dedicated speech-to-text API. If you need to find and quote what was said, read on.

How to do it with Audioscrape

1. Get a key. Sign up free (Google or Microsoft sign-in, no passwords), open your workspace and create a key from the API Keys tab. Send it as Authorization: Bearer YOUR_API_KEY. Reference: Authentication.

2. Search.

curl -X POST "https://www.audioscrape.com/api/search" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "\"interest rates\" AND inflation",
    "search_type": "text",
    "filters": {"speakers": ["Jerome Powell"], "date_range": {"start": "2026-01-01"}},
    "pagination": {"limit": 20},
    "options": {"include_highlights": true, "include_aggregations": true}
  }'

search_type is text, semantic or hybrid. Queries support boolean operators, exact phrases, exclusions and wildcards. Filters accept podcast IDs, speaker names and a date range. Responses include segment.text, segment.start_timestamp, segment.end_timestamp, segment.speaker, the episode and podcast objects, and urls.segment, urls.episode and urls.audio. Aggregations give you counts by podcast, speaker and month. Reference: Search API.

3. Read full transcripts and navigate shows. The MCP tools cover this directly: search_podcasts to find a show, list_podcast_episodes to list its episodes, get_episode_overview for chapters, speakers and entities, and get_transcript for the full speaker-labeled transcript (paged for long episodes; follow coverage.next_start_time). The same tools are callable from any HTTP client as JSON-RPC against https://mcp.audioscrape.com/ with your API key; the MCP overview has a copy-paste curl example, and tools/list returns the current schemas. Interactive REST docs are at /api/docs/swagger.

4. Upload your own audio. POST /api/datasets/{id}/items accepts a multipart file or a JSON audio_url. Poll GET /api/items/{id} until it is published; then it is searchable through the same search endpoint. Reference: Upload API.

5. Or skip the code. If the consumer is an AI assistant, connect it to the MCP server and let it call the tools itself. Claude, ChatGPT, Cursor and any MCP client work; see integrations.

What you get

• Speaker-attributed segments with timestamps and source links for every result.
• Full transcripts with speaker labels, chapters where available, and extracted entities.
• Podcast metadata, charts across 31 countries, trending entities and a knowledge graph of who was mentioned with whom.
• Date-pinned API versioning and a published OpenAPI spec.
• One quota shared between REST and MCP. Free: 100 searches per month (web, REST or MCP) and 30 transcription minutes; the REST data endpoints for libraries, items and transcript export need a paid plan. Starter: 500 searches and 5,000 API calls per month. Pro: 5,000 searches plus 2,000 semantic searches, unlimited API calls. Enterprise: a custom contract with multiple dedicated workspaces, a DPA and region pinning.

Limits and honest caveats

• Not a real-time speech-to-text API. Transcription of uploads is batch and asynchronous.
• Search returns segments, not full episodes; fetch the transcript separately when you need it all.
• The search endpoint is rate limited to 30 requests per minute per IP with a small burst allowance; MCP tool calls have their own per-minute limits. Enterprise customers can ask for higher limits.
• Not every podcast is indexed, and new episodes arrive in batch. Submit missing shows at /submit.
• Speaker names in the public corpus are attributed by an evidence-gated classifier and can be reported and corrected.
• The public corpus comes from open podcast feeds; rights holders can claim or remove content through the standard DMCA process. Uploads must be audio you have the right to use.
• There are no free trials of paid plans. The free plan does not expire and needs no card.

Alternatives

Some podcast networks publish transcripts on their own sites, and some hosting platforms expose transcripts in the feed for their own shows. Those are worth checking for a single show. For search across shows, or for any show that does not publish transcripts, an index like this one is the practical route.

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