Catalyst Labs / Guides
Our Podcast Transcript API on Apify turns podcast episodes into transcripts for AI agents. Give it an RSS feed, an Apple Podcasts show or episode link, an episode web page or a direct audio URL. When the show publishes its own transcript (Podcasting 2.0 podcast:transcript), it returns that text exactly, with speaker labels where the publisher includes them. Otherwise it transcribes the audio with speech-to-text (Whisper large-v3).
Add this MCP server URL to Claude, Cursor or any MCP client (Apify token required):
https://mcp.apify.com/?tools=catalyst-labs/podcast-transcript-api
curl -X POST "https://api.apify.com/v2/acts/catalyst-labs~podcast-transcript-api/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
-H "Content-Type: application/json" \
-d '{"sources":["https://podnews.net/rss"],"maxEpisodesPerFeed":1,"transcriptMode":"auto","outputFormats":["text","segments","markdown"]}'
transcriptMode is auto (publisher transcript first, then speech-to-text), publisherOnly or speechToText. outputFormats accepts text, segments, llmText, srt, vtt and markdown. Cost caps: maxAudioMinutes (default 120) and maxMinutesPerEpisode. onlyNewEpisodes with a monitorId returns only episodes not seen on earlier runs.
On Oct 20, 2026 the release build passed 16 of 16 cloud test cases across 11 podcast hosts, Apple links, episode pages, spending limits and invalid inputs. In each run the charged events matched the delivered records. Speech-to-text word error rate against the show's own published transcript was 2.4% and 3.0% on two English episodes, 10.2% on a Spanish episode and 16.9% on a German one (mostly number formatting). On the same inputs we scored 10 of 10 on our feature checks; the most-used alternative scored 6.
The most-used alternative is cheaper per speech-to-text minute ($0.006) and slightly more accurate on one English clip (2.1% vs 3.0%). Speech-to-text output has no speaker labels. Spotify-only links are not supported.