Video to Markdown: What Actually Works (and What Doesn't)

"Video to markdown" is a search people make with one of two very different jobs in mind. I know because I tried both before I understood the difference.

Job one: you want the transcript of a video as clean Markdown, so you can paste it into ChatGPT or Claude and ask for a summary, a set of quotes, or a study outline.

Job two: you want the page the video lives on as Markdown — the title, the description, the chapter list, the top comments, the course notes beside the player — because that surrounding text is the actual content you care about.

Most tools only do one of these, and most blog posts about this topic do not tell you which. So let me be direct about what I found.

The short version

• A video file itself is not text. Nothing converts an MP4 to Markdown without a speech-to-text step somewhere.
• If the video already has captions (YouTube auto-captions, Vimeo, Loom, most course platforms), the text is already sitting in the page or one HTTP request away. You do not need a transcription service.
• If the page requires a login — a paid course, an internal Loom, a company all-hands on a private channel — server-side converters cannot see it at all.

That last point is where I spent the most time, so it gets the most space below.

What I tested

I ran four tools against the same set of pages: a public YouTube video with auto-captions, a Loom shared via unlisted link, a paid course lesson on a platform I subscribe to, and a conference talk page with a transcript tab.

Jina Reader (r.jina.ai/<url>) is the fastest thing to reach for. Prefix a URL, get Markdown back. It is genuinely good, free for light use, and I still use it for public pages. On the YouTube page it returned the title, description, and metadata cleanly. On the paid course lesson it returned the login wall, because it fetches the page from its own servers with no session of mine.

Firecrawl is the more serious version of the same idea — it renders JavaScript, handles pagination, and has a real API. If you are building a pipeline that crawls hundreds of pages, this is the right tool and I am not going to pretend otherwise. It also has a free tier, then paid plans, and it needs an API key. Same structural limit: it is a server fetching a URL, so authenticated pages are out unless you hand it cookies, which I would rather not do.

MarkDownload is a browser extension, which means it reads the page you are actually looking at, session and all. It uses Readability under the hood and produces solid article Markdown. On video pages it was hit or miss — Readability is tuned for articles, and a YouTube watch page is not an article, so I often got the description and little else.

Web2MD is what I build, so take the bias as stated. It is a Chrome extension, so like MarkDownload it converts the DOM in front of you rather than refetching the URL. The practical difference on video pages: it keeps the structural text around the player — description, chapter headings, transcript panel when one is open — instead of discarding everything Readability does not score as article body.

The login problem, concretely

Here is the test that decided it for me. I opened a lesson page on a course I pay for. The lesson has a video plus a long written transcript tab below it.

Through Jina Reader, I got this:

# Sign in to continue

Enter your email address to access your courses.

[Sign in](https://example-course.com/login)
[Forgot password?](https://example-course.com/reset)

That is the correct behavior. The server is not me, and the page is behind my account. No amount of configuration changes that without handing over credentials.

With the transcript tab open in my own tab and Web2MD run on it, I got the actual lesson:

# Lesson 4: Handling Backpressure in Streams

## Transcript

So in the last lesson we built a reader that pulls as fast as
it can. That works until the consumer is slower than the
producer, and then your memory usage climbs until something
falls over.

The fix has a name: backpressure. The consumer tells the
producer to slow down.

## Chapters

- 00:00 - Why the naive reader breaks
- 02:14 - What backpressure means
- 07:40 - Implementing pause and resume
- 14:02 - Testing under load

**Tokens: 1,847**

That token count at the bottom is there because I got tired of pasting something into Claude and discovering halfway through that it was 40k tokens. A long lecture transcript will blow past a context window quietly. Seeing the number before you paste saves a round trip.

How to actually get a video transcript into Markdown

Practical steps, in order of how often they work:

1. Open the transcript panel first. On YouTube, click the three dots under the video, then "Show transcript." The text is now in the DOM. Any browser-side converter can pick it up. Server-side tools cannot, because the panel loads only after the click.
2. Check for a transcript tab. Most course platforms, Loom, and conference sites have one. Same principle — open it, then convert.
3. If there are no captions at all, you need real speech-to-text. Whisper locally, or a paid transcription API. No Markdown converter will invent text that is not there, and anything promising otherwise is converting the captions and not telling you.

Step 1 is the one people miss. The reason "video to markdown" feels broken with server-side tools is usually that the transcript exists but is not in the initial HTML.

Honest limits

Web2MD is Chrome only right now. No Firefox, no Safari. If you are not in Chrome or a Chromium browser, use MarkDownload.

The free tier is 3 conversions per day. That is enough to try it and enough for occasional use, and it is not enough for daily work. Pro is $9 per month for unlimited. I would rather say the number than make you dig for it.

It does not transcribe audio. If a video has no captions, Web2MD gives you the page text and nothing more. That is a real gap and I am not going to dress it up.

And if your job is crawling a thousand public video pages unattended, a browser extension is the wrong shape of tool. Use Firecrawl. Browser-side conversion means a browser has to be open, with you in it.

Where each one wins

Server-side readers win on public pages at volume and on automation. Firecrawl in particular is built for that and does it well.

Browser-side conversion wins when the content needs your session: paid courses, internal Looms, private Slack-embedded videos, anything behind SSO. It also means the page content never leaves your machine on the way to a conversion server — for internal company videos, that matters more than speed.

If you want the broader comparison beyond video pages, I wrote up the full web page to Markdown workflow and a more detailed Jina Reader comparison covering the same tradeoffs on regular articles.

Try it on one page

Pick the video page that annoyed you most recently — the one where you wanted the transcript in Claude and ended up copy-pasting in chunks. Open the transcript panel, run a conversion, check the token count before you paste.

Web2MD is on the Chrome Web Store with 3 free conversions a day, no account and no API key. If it does not handle your page, the free tier costs you nothing to find out.