Frequently asked questions

SubDownload is an AI YouTube knowledge base. Paste any YouTube link, and you get a clean transcript, an instant AI summary, and a searchable note saved to your personal Knowledge Base. It also exposes that backend through a REST API, an MCP server, and Agent Skills, so AI agents like Claude, ChatGPT, and Cursor can read YouTube directly.

Yes — there's a free tier you can use without creating an account. Paste a YouTube URL and the basic transcript and summary work right away. Heavier usage and the developer integrations (MCP, REST API) require signing in for an API key.

Open subdownload.com, paste the video URL into the input box, and the transcript renders alongside the player within seconds. You can copy the full transcript with one click, or export it as txt, srt, vtt, csv, json, or md.

Yes. When YouTube hasn't generated captions for a video — Loom recordings, niche channels, fresh uploads, or talks where captions were disabled — SubDownload runs an AI transcription pass to produce text. The same paste-link → transcript flow works either way.

After a transcript is available (either from YouTube's captions or from AI transcription), the text is fed into an LLM that produces a structured summary: a short overview, a section-by-section outline, and quotable spans with timestamps you can click to jump to. The summary lands in seconds, so you can decide whether the full video is worth a real watch.

Yes. SubDownload exposes an MCP server at api.subdownload.com/mcp that plugs into Claude Desktop, Claude Code, Cursor, and other MCP-compatible clients. Your AI agent can then fetch transcripts, list channel videos, and summarize on demand, without you copy-pasting between tabs. There is also a REST API and an Agent Skills bundle (npx @subdown/skill@latest) for clients that prefer those paths.

MCP (Model Context Protocol) is the spec that lets a Claude or Claude-compatible client call out to external tools. Without something like MCP, an AI agent can't read YouTube — it has no browser, no audio decoder, and no caption API. An MCP server bridges that gap: the agent says 'fetch this transcript' or 'search this channel' and the server does the work and returns text the model can reason over.

Source-language transcripts are supported for the languages YouTube and the AI transcription layer can handle (covering most major world languages). Translation is available for cross-language workflows. The product UI itself ships in 11 languages: English, Chinese, French, Russian, Arabic, Spanish, German, Portuguese, Japanese, Hindi, and Korean.

Six formats: plain text (txt), SubRip (srt), WebVTT (vtt), comma-separated values (csv), JSON, and Markdown (md). srt and vtt include timestamps for use in subtitle workflows; csv and json are for piping into other tools; md is convenient for direct paste into note apps like Obsidian or Notion.

Otter.ai is built around live meeting transcription; it doesn't natively pull arbitrary YouTube videos into a personal library. NotebookLM is a notes-first AI app where you paste sources for ad-hoc Q&A, but it's not a YouTube-native pipeline with a long-running searchable library plus an MCP server for agents. SubDownload is YouTube-first, with a personal knowledge base and an MCP server as core surfaces, so it composes with your existing AI workflow rather than replacing it.

No. The basic transcript and summary flow on subdownload.com works without an account. Sign up once if you want to save videos to your personal Knowledge Base, get an API key for developer access, or use the MCP server.

Email contact@subdownload.com — a real person reads every message. Bugs, feature requests, integration questions, and pushback on design decisions are all welcome.