My reading list used to die in browser tabs. Dozens of them, opened with the best intentions, never read, and impossible to search later. Bookmarks were not much better — a title and a URL tell me nothing about what was actually in the article.
So I started filing articles as Markdown notes instead. Plain text files, named properly, greppable, diff-able, and readable by every tool I use — including the AI assistants I ask questions about my own notes. The only tedious part was the conversion: copy-paste mangles tables, drops headings, and drags half the page's navigation along with it.
Now I do it with one API call.
The whole workflow is one call
I keep a plain list of URLs I want to keep. When it grows, I send the whole list to a batch endpoint:
curl -X POST "https://page2md-api.pages.dev/convert/batch" \
-H "Content-Type: application/json" \
-d '{"urls":["https://en.wikipedia.org/wiki/Markdown","https://example.com","https://en.wikipedia.org/wiki/Hyperlink"]}'
That is a real call I ran today, with a real response. The summary line:
"summary": {"total": 3, "success": 2, "failed": 1}
The Wikipedia article on Markdown came back as 45,011 characters of clean Markdown — headings, links and structure intact. The Hyperlink article came back at 38,642 characters. Here is the actual opening of the first note:
Markdown is a lightweight markup language for creating formatted text using a plain-text editor. John Gruber created Markdown in 2004 as an easy-to-read markup language. Markdown is widely used for blogging …
Each result also carries the page title, byline and excerpt, so my filing script can name the note and write a one-line summary at the top without me opening the file.
A reading list is a batch, not a loop
Two details make the batch endpoint the right shape for this job:
- One request for the whole list. Up to 25 URLs per call. No loop, no per-URL bookkeeping in my script — I hand over the list and file whatever comes back.
- Failures are per-item, not per-batch. Which brings me to the third URL in that call.
The honest failure is the feature
example.com is a placeholder page with a couple of lines of text. The API did not pad it, scrape the chrome around it, or hand me a near-empty file with a success flag. It returned this, inside the same response:
{"url":"https://example.com","code":"CONTENT_TOO_SHORT","error":"Extracted content is too short or empty"}
That one behavior is why I trust the pipeline with a whole list. Pages with too little real content say so, in machine-readable form, and the rest of the batch still completes. My notes folder only ever gets files that are worth searching later.
Why Markdown specifically
HTML is for browsers. Markdown is for me: it reads fine raw, renders fine everywhere, versions cleanly in git, and drops straight into any notes app or LLM context window without a parsing step. Converting at capture time — once, when the article enters my system — beats re-extracting it every time I want to use it.
Try it without writing code
If you want to see what a page looks like as clean Markdown before wiring anything up, there is a free web tool that runs the same conversion in the browser: https://page2md-api.pages.dev/url-to-markdown — paste a URL, get the Markdown. My reading list has been empty-tab-free ever since.
Top comments (0)