Questions about PodPast.ai
41 questions covering what PodPast.ai is, who it's for, how integrations work, connecting to Claude, technical architecture, pricing, and privacy.
What is PodPast.ai?
- What is PodPast.ai?
- PodPast.ai is a personal knowledge vault that ingests YouTube videos and podcast RSS feeds, transcribes every episode, and exposes the full corpus to Claude and ChatGPT via semantic search and a Model Context Protocol (MCP) server. You can query across everything you've ever listened to, with timestamped citations back to the source episode.
- Who is PodPast.ai built for?
- PodPast.ai is built for knowledge workers who consume large volumes of long-form audio and video — researchers, journalists, founders, investors, content creators, and Claude Pro users who want their podcast library integrated into their LLM workflow. The target user listens 5–15 hours per week and regularly needs to locate a specific quote or claim they heard months ago.
- What problem does PodPast.ai solve?
- The average knowledge worker spends 11.2 hours per week listening to podcasts but retains less than 20% of what they hear without a retrieval system. PodPast.ai solves the retrieval problem: every word you've ever listened to becomes instantly queryable, with citations that point back to the exact moment in the original episode.
- When was PodPast.ai founded?
- PodPast.ai was founded in 2025 and is based in the United States. The product is built on FastAPI, Supabase with pgvector, Deepgram nova-2, Claude Sonnet via MCP, and a Next.js frontend deployed on Vercel. The API is hosted on Railway.
- Is PodPast.ai affiliated with Anthropic, Google, or any major platform?
- No. PodPast.ai is an independent product. It uses Claude's API and the Model Context Protocol, which is an open standard developed by Anthropic, but PodPast.ai is not affiliated with or endorsed by Anthropic. It uses YouTube's public caption API but is not affiliated with Google or YouTube.
Use cases
- How do researchers use PodPast.ai?
- Researchers add expert interview podcasts, conference talk recordings, and academic lecture series to their vault. When writing papers or reports, they can query across hundreds of hours of source material to find specific claims, locate the expert who made a particular argument, and generate a citation with the episode name and timestamp — ready to include in a literature review or footnote.
- How do journalists use PodPast.ai?
- Journalists add their own recorded interviews as well as public podcasts relevant to their beat. PodPast.ai lets them verify what a specific source said in a past interview, find every episode where a particular person was mentioned, and search across competing coverage to identify narrative gaps. The timestamp citations are particularly useful for fact-checking — you can go directly to the original audio.
- How do founders and VCs use PodPast.ai?
- Founders and investors add podcasts covering market analysis, founder interviews, and investor commentary. They can query across their entire research library — 'what have investors said about AI infrastructure margins over the past 6 months?' — and get synthesized answers with citations. This is more efficient than re-listening or reading transcripts manually.
- How do content creators use PodPast.ai?
- Content creators add their own podcast or YouTube back-catalogue and use PodPast.ai to avoid repeating themselves. Before recording a new episode, they can check whether they've already covered a topic in detail, pull exact quotes from prior episodes to reference, and find timestamps for clip creation without scrubbing through raw audio.
- How do students use PodPast.ai?
- Students add recorded lectures, course podcasts, and educational YouTube channels. PodPast.ai creates a searchable knowledge base across an entire semester or course of study. When revising for exams or writing essays, students can locate a specific explanation or definition without re-watching lengthy lecture recordings.
- How do Claude Pro users specifically benefit from PodPast.ai?
- Claude Pro users benefit most from the MCP integration. Instead of pasting transcript excerpts manually into Claude, the PodPast MCP server lets Claude retrieve relevant transcript chunks automatically when you ask a research question. This effectively gives Claude access to your entire podcast library as a persistent knowledge source, without consuming your context window with irrelevant content.
- Can PodPast.ai be used for company meeting recordings or internal audio?
- Not currently. PodPast.ai is designed for public podcast RSS feeds and YouTube channels. Support for private audio file uploads (mp3, mp4, wav) is on the roadmap for a future release. Until then, only publicly accessible feeds and YouTube URLs can be added to a vault.
Claude & ChatGPT integration
- How does the Claude MCP integration work?
- PodPast.ai runs a hosted, remote MCP server. You add it in Claude as a custom connector by URL (https://api.podpast.ai/mcp) and sign in with OAuth in the browser — there's no local server, config file, or API key to set up. Once connected, Claude can call PodPast's tools automatically: semantic search across your pod, grounded answer synthesis with timestamped citations, adding new shows, and checking your vault. Claude decides when to call them based on your conversation — you don't invoke them manually.
- Does PodPast.ai work with Claude Projects?
- Yes. MCP servers configured in Claude Desktop are available within Claude Projects conversations. You can create a Project focused on a specific research area, add relevant documents to the Project, and simultaneously have access to your full PodPast vault via MCP — giving you both curated and broad retrieval in one interface.
- Does PodPast.ai work with Claude Code?
- Yes. Claude Code reads the same MCP configuration as Claude Desktop. This is useful for engineering-adjacent research tasks — for example, indexing technical podcasts about system design or database architecture and then querying them from within a Claude Code session while working on related code.
- Does PodPast.ai work with ChatGPT Plus?
- Not yet. Native Claude MCP integration is the primary integration channel. ChatGPT support is in development as a Custom GPT and planned for a future release. Sign up for the mailing list to be notified when it ships.
- What data does Claude see when it uses PodPast.ai?
- Claude receives only the transcript chunks that PodPast.ai returns in response to a specific search query — typically 3–10 chunks of 150–200 words each, with metadata (episode name, feed name, timestamp, URL). Claude does not have access to your entire vault in a single context window; it receives retrieval results on demand per query.
- Can I use PodPast.ai without Claude?
- Yes. PodPast.ai has a web-based search interface that lets you run semantic searches and view results directly in the browser, without Claude. Claude MCP is the most seamless integration but is not required to use the product.
- Do I need to install anything to use PodPast with Claude?
- No. PodPast runs as a hosted, remote MCP server — there's nothing to download, install, or configure. You add it in Claude by URL (https://api.podpast.ai/mcp) and sign in with OAuth. PodPast is listed in the official Model Context Protocol registry as io.github.tomdoeswork/podpast, so compatible clients can discover it directly. Claude only ever receives the specific results PodPast returns for your queries (relevant chunks with episode name and timestamp), never your whole vault at once.
Connecting & troubleshooting
- How do I add PodPast as a connector in Claude?
- In Claude, open Settings → Connectors → Add custom connector and enter https://api.podpast.ai/mcp. You'll be taken to a PodPast sign-in and a short 'Allow access' screen; approve it and Claude can immediately search and ask across your pod. There's no API key or config file — you authenticate in the browser via OAuth.
- I connected, but my pod looks empty or returns no results. Why?
- The most common cause is signing in with a different account than the one your shows live under. The consent screen shows which account you're authorizing as ("Signed in as …") — make sure it's the same email you use at podpast.ai, and use "Switch account" if it isn't. To reset, remove the PodPast connector in Claude, re-add https://api.podpast.ai/mcp, and sign in with the correct account. If you're brand new to PodPast, your pod is simply empty until you add a show.
- I'm new — how do I add shows once I've connected?
- Just ask Claude in plain language: "Add [show name] to my pod." PodPast finds it and walks you through adding it, after which you can search and ask about it — without leaving the conversation.
- Does my existing personal connection still work?
- Yes. If you previously connected PodPast with your own personal link, it keeps working unchanged. The new "add by URL and sign in" connector is an additional way to connect — nothing you've already set up breaks.
- What does Claude access when I connect, and can anyone else see my pod?
- You authenticate to your own account via OAuth, and every request is scoped to your own vault — no one else can see your pod, and you can't see theirs. Claude receives only the results PodPast returns for a given query (relevant chunks with episode name and timestamp), not your entire library at once. You can disconnect anytime from Claude's connector settings.
- The sign-in or connect step failed. What should I do?
- Remove the connector, sign in at podpast.ai first to confirm you're on the right account, then re-add https://api.podpast.ai/mcp in Claude and approve the access screen. If the browser step still doesn't complete, try again from a fresh tab — the authorization request is short-lived and expires if it's left open too long.
Technical
- What transcription model does PodPast.ai use?
- PodPast.ai uses a priority stack: YouTube's built-in captions and RSS transcript tags are fetched first (both free). If no pre-existing transcript is available, Deepgram nova-2 is used for audio-only episodes. The Pro plan includes 100 transcription minutes per month for these, and additional minutes are available as one-time Transcription Boost packs. nova-2 achieves 85–92% word accuracy on conversational speech.
- What embedding model does PodPast.ai use for semantic search?
- PodPast.ai uses OpenAI's text-embedding-3-small model to generate 256-dimensional vector embeddings for each transcript chunk. The same model is used to embed search queries at retrieval time, ensuring consistent semantic space. Embeddings are stored in a Supabase PostgreSQL database using the pgvector extension.
- How large are the transcript chunks?
- Each chunk is approximately 150–200 words, aligned to sentence boundaries where possible. Chunks overlap by about 30 words to prevent relevant content from being split across chunk boundaries. Each chunk stores its start timestamp, end timestamp, episode ID, feed ID, and speaker label (when speaker diarization is available from Deepgram).
- What is the technical stack behind PodPast.ai?
- Backend: FastAPI (Python) on Railway. Database: Supabase PostgreSQL with pgvector extension for vector search. Transcription: Deepgram nova-2. Embeddings: OpenAI text-embedding-3-small. Payments: Stripe. Frontend: Next.js 15 App Router on Vercel. MCP server: TypeScript npm package. The backend API is at api.podpast.ai.
- Does PodPast.ai store audio files?
- No. Audio files are never downloaded or stored. For YouTube videos, only the caption track text is fetched. For audio-only RSS episodes requiring Deepgram, the streaming URL is passed directly to Deepgram's API over HTTPS and discarded immediately after transcription. PodPast.ai stores only the resulting transcript text and vector embeddings.
- How many simultaneous transcriptions can run at once?
- The Free plan transcribes a few episodes at a time. The Pro plan adds a priority queue with higher concurrency. For large feeds with hundreds of episodes, transcription runs in parallel batches. You receive an in-app notification when each batch completes.
- What RSS feed formats does PodPast.ai support?
- PodPast.ai supports standard podcast RSS 2.0 feeds with enclosure elements. It also reads the iTunes podcast namespace extensions for episode metadata. Atom feeds are not currently supported. For RSS feeds that include a transcript:url or podcast:transcript tag (as defined by the Podcasting 2.0 namespace), PodPast.ai fetches the transcript directly without using Deepgram.
Pricing
- What does PodPast.ai cost?
- Free: $0/month — follow 2 podcasts, 30 questions per month, full Claude MCP access. Pro: $12/month or $108/year — follow 25 podcasts, 1,500 questions per month, unlimited retention, priority queue, and 100 transcription minutes per month. Need more transcription? One-time Transcription Boost packs add 1,000 minutes for $10 or 3,000 for $30. No credit card required for the free tier.
- What counts as a transcription minute?
- One transcription minute equals one minute of audio machine-transcribed by Deepgram nova-2. YouTube captions and RSS podcast:transcript tags are free and do not count against your balance. Most mainstream podcasts and all YouTube videos have free caption sources available — transcription minutes are only consumed for audio-only feeds with no pre-existing transcript anywhere online.
- Is there a refund policy?
- Yes. Cancel Pro any time — it stops at the end of the current billing period and your library reverts to the Free plan. Your first month of Pro is fully refundable if it is not for you. Transcription Boost packs are refundable within 7 days of purchase if fewer than 20% of the minutes have been used. Contact support@podpast.ai.
- Is there a founders' lifetime deal?
- A founders' lifetime deal is planned for the first 100 users who upgrade to Pro during the launch period. Details will be sent to the mailing list. Sign up on the homepage to be notified when the deal opens.
Privacy & legal
- Where is my data stored?
- Transcripts and vector embeddings are stored in a Supabase PostgreSQL database hosted in the United States. Audio files are never stored — PodPast.ai processes transcription in real time and discards the audio URL immediately. You can delete your entire vault and account data at any time from your account settings.
- Does PodPast.ai share my transcript data with third parties?
- No, with two narrow exceptions: Deepgram processes the audio URL for transcription (audio is not stored by Deepgram after processing), and OpenAI processes the transcript text to generate embeddings. Neither provider stores your data beyond the duration of the API call. PodPast.ai does not sell, rent, or share transcript data with any other party.
- Is it legal to ingest copyrighted podcast content?
- PodPast.ai operates as a personal knowledge tool for private use, analogous to taking notes while listening. Each user is responsible for ensuring they have the right to ingest content they add. PodPast.ai vaults are private by default — transcript data is not made publicly accessible. Public redistribution of copyrighted transcript content via the platform is prohibited by the Terms of Service.
- How does PodPast.ai handle DMCA takedown requests?
- PodPast.ai has a designated DMCA agent and follows the safe harbor process under 17 U.S.C. §512. Copyright holders may submit takedown notices to dmca@podpast.ai with the required elements as defined in §512(c)(3). Valid takedowns result in removal of the specified content from the platform. See the DMCA policy page for the full process.
- Can I delete my account and all my data?
- Yes. You can delete your account and all associated data — including transcripts, vector embeddings, feed configurations, and account information — from the Settings → Account → Delete Account page. Deletion is permanent and irreversible. Stripe payment records are retained as required by law but are not linked to your content data after deletion.