You've listened to hundreds of hours. You still can't search a word of it. Spoken returns an episode — or a show's whole back catalog — as Markdown with real speaker names, ready for your notes or your LLM.
"I used Spoken to add every My First Million episode to my knowledge base, with a cron to pull new ones. Now I can enjoy the podcast on a run, then chat with Claude about it later — every episode saved and accessible in my Claude sessions."
— Marcus Taylor
17,500+ transcripts fetched by people building searchable podcast archives (September 2026)
Two API calls. Search by text or paste a URL from Spotify, YouTube, or any podcast app — then fetch the full transcript.
# 1. Search for an episode
curl -H "x-api-key: pt_demo" \
https://spoken.md/search?q=huberman+sleep
# 2. Get the transcript
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Response is text/markdown with speaker names and credit info in headers:
**Andrew Huberman** (0:00)
Welcome to the Huberman Lab podcast,
where we discuss science and science-based
tools for everyday life. Today my guest is
Dr. Matt Walker, professor of neuroscience
at UC Berkeley and author of Why We Sleep.
**Matt Walker** (0:45)
Thank you for having me, Andrew. Sleep is
one of those things where small changes to
your routine can have outsized effects on
both mental and physical health.
...A show you follow has a run — sometimes hundreds of episodes, searchable once they're sitting in your notes or knowledge base. List them in one call, then loop them through the transcript endpoint:
# List every episode of a show (free — metadata only)
curl -H "x-api-key: pt_demo" \
https://spoken.md/podcasts/{podcastId}/episodesReturns a JSON episodes array with an id for each — fetch them through /transcripts/{id} to build the full archive. Every /search result carries a podcastId, so you can pivot from any episode straight to its whole show.
Listing episodes costs nothing — you're charged only for transcripts you successfully fetch. A 300-episode back catalog is 300 credits: $30 at the 500-pack rate. Point the same call at a cron afterwards and new episodes land in your archive as they publish.
Most people start with the one show they keep quoting. Then they add the rest of the rotation. It's the same two calls per show — list the episodes, loop them through the transcript endpoint — and the shows you actually listen to end up in one place, as text you can ask questions of.
Search it and Spoken counts the episodes it can reach, then prices the whole run on the show's page — one credit per episode, and nothing for the rest. Marcus pulled all 884 episodes of My First Million that way, then set a cron for the new ones.
"Podcasts are where frontier tech gets discussed first — often well before it reaches papers or press. That makes podcast ingestion critical infrastructure for my research agents, not a nice-to-have. Spoken gave me a comprehensive transcript library in one place and saved me days of standing up my own transcription stack. Because it's priced per API call rather than by subscription, I didn't have to commit to a year to find out whether it worked. I'd been evaluating options for a while — this was exactly what I was looking for."
— Kydo, Eigen Labs
Doing this yourself for one episode is an afternoon. Doing it for a show's back catalog is a project — and speech-to-text APIs bill per minute of audio, so a few hundred episodes puts hundreds of hours on the meter before you've written a line of code to work out who's talking.
Spoken bills per episode, not per minute. One call returns speaker-labeled Markdown with real names, ready to drop straight into your notes or your LLM's context window. The back catalog stops being a project and becomes a loop.
Figuring out who is speaking is harder than transcribing what they say. Most tools punt — you get "Speaker 1," "Speaker 2," and a manual cleanup job. Spoken resolves speakers to their real names from context.
Speaker 1 (0:00)
Welcome to the podcast, where we discuss science and science-based tools for everyday life.
Speaker 2 (0:45)
Thank you for having me. Sleep is one of those things where small changes can have outsized effects.
**Andrew Huberman** (0:00)
Welcome to the podcast, where we discuss science and science-based tools for everyday life.
**Matt Walker** (0:45)
Thank you for having me. Sleep is one of those things where small changes can have outsized effects.
No post-processing, no manual cleanup, no guessing.
That is the claim, and you do not have to take it on trust: search one of your own shows and open an episode. You will see a real excerpt with whatever names Spoken resolved — free, no key, no card.
Seen enough? Your key is ready in 30 seconds.
# Agent skill (Claude Code, Cursor, Windsurf, etc.)
npx skills add https://spoken.md
# OpenAPI spec — works with any agent framework
https://spoken.md/.well-known/openapi.json
# llms.txt — automatic LLM discovery
https://spoken.md/llms.txt
# MCP server (Claude Desktop, Cursor, Cline, etc.)
npx spoken-mcp$0.15 per transcript
$0.10 per transcript save 33%
$0.08 per transcript save 47%
For scripts & back-catalogs
One credit = one episode. A typical back catalog runs around 250 episodes — 500 covers about two shows, 2,000 covers a shelf. Long-running shows go well past that, so price yours before you pick: search your show.
Credits never expire. Errors are never charged. And if Spoken isn't what you expected, email us within 14 days with 20 or fewer credits used and we'll refund the pack in full — see Refunds.
Prices exclude tax. Tax is added at checkout based on your location. Stripe is the merchant of record for the purchase.
Once you have a key, top-ups cost less — 100 for $10, 500 for $40, 2,000 for $120, down to $0.06 each. The price depends only on how many you buy.
Each transcript replaces roughly a dollar of transcription infrastructure. No subscription, no overage charges, no annual commitment.
Already have a key? Top up here, check your balance and usage, or let your agent top up automatically via the API.
Spoken serves the shows that publish a transcript. Search yours — it is free, it takes five seconds, and every episode it returns is one you can fetch.
Search by text or paste a URL from Spotify, YouTube, or any podcast app. Shows in other languages work too — Italian, German, Spanish, French and Portuguese among them — and each transcript comes back in the language it was recorded in, never translated. Availability varies episode by episode rather than by language.
Yes, and you don't need to hunt down each episode by hand. One call to /podcasts/{podcastId}/episodes lists every episode of a show that has a transcript available, and that call is free — it returns metadata only. Fetch them by ID from there: one credit each, and re-fetching an episode you already pulled is never charged again.
Speaker names are detected automatically by analyzing the transcript for name mentions — no manual lookup table or post-processing required. When real names cannot be determined from context, labels like "Host" or "Guest" are used as fallbacks.
Transcripts are returned as clean Markdown with speaker names in bold and timestamps per turn. A typical one-hour podcast episode produces 8,000–15,000 tokens — sized to fit in most LLM context windows in a single call. No proprietary markup, no timing artifacts, no post-processing needed.
Yes. Install the agent skill with npx skills add https://spoken.md, add it as an MCP server with npx spoken-mcp (Claude Desktop, Cursor, Cline, etc.), use the OpenAPI spec for any agent framework, or make a plain HTTP call — anything that can hit an API works.
Yes, two of them, and neither asks for a card.
To judge the transcripts: search your show and open any episode. You get a real excerpt of that episode with real speaker names — the actual output, on the show you care about, rather than a sample we picked.
To judge the API: the demo key pt_demo works on /search and /podcasts/{id}/episodes for any show, and returns one full transcript for a sample episode, so you can see the exact response shape and headers before buying.
Paste your API key in the top-up form above, or let your agent handle it — the API response includes a top-up link when credits run out.
No. One credit per episode, whatever its length — a four-hour interview costs the same as a twenty-minute one. Nothing is billed per minute of audio, so a show with long episodes costs no more than a show with short ones.
Transcripts start at $0.15 each (100-pack for $15), with volume discounts: 500 for $50 ($0.10 each) or 2,000 for $160 ($0.08 each). Returning customers get lower top-up rates. Errors (404, 502) are never charged. No subscription — credits never expire.
No — listed prices exclude tax. Tax is added at checkout based on your billing location: EU customers pay their own country's VAT rate, so a $50 pack comes to $60 in France, and UK customers pay 20%. EU businesses that enter a valid VAT number pay no VAT. Where no tax applies you pay the listed price. Stripe is the merchant of record and emails an invoice after every purchase.
Browse transcripts from popular shows:
Example transcripts by podcast → · Browse by topic →
From $0.08/transcript. No subscription, no expiry. Try with pt_demo first.