Turn a YouTube Playlist Into a Queryable Course

A playlist is a syllabus with no index

We re-read BrainTube's own connect page and Recall's public MCP documentation on 4 September 2026, looking for the shortest honest path from a saved YouTube playlist to an answer you can act on. A conference track, a lecture series, a maintainer's playlist of release walkthroughs: each is already a course, and what none of them has is an index. You can watch them in order and scrub a progress bar, but you cannot ask the set of them a question and be pointed at the place where it is answered.

That is the gap this walkthrough closes. Playlists are the unit here: one playlist, connected once, then queried.

Connecting the playlist

The import path is a playlist URL. BrainTube lists YouTube playlists among its supported imports, alongside web articles, PDFs and EPUBs, Notion pages, Markdown and text files, and screenshots transcribed to text. There is also a "Liked Videos" sync on the YouTube readonly scope, useful if your habit is to like a talk while watching it rather than to organise a playlist afterwards.

How much comes in at once depends on your plan. The bulk YouTube import caps are 10 items on Free, 50 on Starter, 150 on Pro and 400 on Pro+ (pricing). Those caps are one of only two server-enforced tier gates in the product; the other is Proactive AI Insights, on Pro and Pro+. Everything else is metered rather than gated. Saving costs nothing and earns +1 BrainToken, so the import is not the expensive part of this exercise.

A 30-video conference track lands in one pass on Starter. A 200-video lecture series on Free does not, and that is worth knowing before you start rather than after.

From transcript to first citation

Once the videos are in, the transcripts are what get indexed, not the titles. That is the difference between a playlist you have bookmarked and one you can interrogate.

The first question worth asking is a narrow factual one, because it is the one you can check yourself. Something like: where does anyone in this playlist define the retry semantics? Over MCP, search_knowledge is the everyday retrieval tool and deep_search the heavier one. The answer comes back as the passage that matched, attributed to the video it came from — which is what makes it verifiable. You go and watch that bit. Either the passage says what the answer said it says, or it does not, and you have learned something about how far to trust the index.

If you already know which talk you want, search_by_source confines the search to one video. If you want the neighbourhood around a result rather than the result alone, get_related walks the entity-level knowledge graph out from it — a Postgres graph with PageRank retrieval, so "related" means related by structure, not by keyword overlap.

All of this runs over one read-write MCP server at https://mcp.brain-tube.com/mcp, publishing 44 tools today. Claude.ai connects by OAuth with no API key, and MCP access is on every plan including Free (connect).

The questions worth asking a whole playlist

Here is the part people underestimate. Having connected a playlist, the instinct is to ask it to summarise itself. That is the least interesting thing a corpus of forty talks can do for you, and it is what you could have got from reading forty descriptions.

The questions that repay a playlist are the cross-cutting ones, expensive precisely because no single video contains the answer:

• Where do two speakers disagree? Two people on the same track, three sessions apart, giving opposite advice on the same decision. Nobody watching in order notices, because the contradiction is separated by four hours of viewing.
• Which claim keeps recurring? A claim that appears once is a speaker's opinion. The same claim in six talks by six people is the field's consensus, and warrants different treatment.
• What does nobody address? This query only makes sense against a bounded set. detect_gaps exists for exactly this shape of question: not what is in the playlist, but what a playlist on this subject conspicuously fails to cover.

That last one is the honest argument for treating a playlist as a course rather than a queue. A queue is something you finish; a course is something you have a map of, including a map of its blank regions.

Two other things then become available on the same corpus. FSRS spaced repetition schedules the claims you decided were worth retaining. Content Lab turns the set into podcasts, digests or quizzes, which is a reasonable second pass over a series you are trying to learn rather than merely reference.

Recall is worth naming here, because their users have asked for the summarising half of this out loud: bulk summaries from a YouTube playlist sits second on their public feature-request board, with 388 votes, marked under review. Their Plus-tier bulk import ceiling is also considerably higher than ours, at up to 1,000 bookmarks or 10,000 Markdown notes (their pricing page, read 4 September 2026). The difference falls on what your assistant can do afterwards: their own developer documentation states that "The MCP server currently provides read-only access to your knowledge base. Write capabilities are coming soon."

Where BrainTube stands

What it does on this topic: imports YouTube playlists and syncs "Liked Videos" on the readonly scope; indexes the transcripts; exposes retrieval over one read-write MCP server that any MCP client can connect to, on every plan including Free; and offers search_knowledge, deep_search, search_by_source, get_related and detect_gaps as the tools you would actually reach for against a playlist.

What it does not do. The bulk import cap is real and low at the bottom of the range: 10 items on Free will not take a conference track, and you will be importing in passes until you are on Starter or above. The "Liked Videos" sync is not watch history and is not Watch Later — if your record of what you have consumed lives in either of those, BrainTube cannot see it. There are no native iOS or Android apps; it is a PWA you add to your home screen, which matters if your intended posture for a lecture series is a phone on a train. And BrainTokens are spent on AI actions, at the same per-action cost on every plan. Saving is free and earns +1; asking is not free. If the plan is to put fifty questions to a 150-video corpus in an evening, budget for it.

We have written the import side of this up separately in bulk import a YouTube playlist into a knowledge base.

How these figures were checked

BrainTube's tool count, tool names, import support and MCP endpoint were read from https://brain-tube.com/connect.md on 4 September 2026, HTTP 200. Plan prices, BrainToken allowances and the bulk-import caps were re-confirmed on https://brain-tube.com/pricing the same day, HTTP 200. Claims about BrainTube describe functionality running in production on that date.

Recall's tier contents, including the Plus bulk-import ceiling of 1,000 bookmarks or 10,000 Markdown notes, were read from https://www.recall.it/pricing on 4 September 2026, HTTP 200. The read-only MCP quotation is verbatim from https://docs.recall.it/developer/mcp, read the same day.

One figure carries a caveat. The 388-vote count and the "under review" status for bulk summaries from a YouTube playlist are as of the packet date, 4 September 2026, not re-verified live: Recall's feature-request board is a Canny instance that renders its post list and vote counts in JavaScript, so the 4 September fetch returned the shell and navigation only. The vote totals come from a reading dated 25 August 2026. Treat the number as approximately right and the ranking as the more durable fact.

If you spot something here that has gone stale, tell us and we will correct it with the date attached.