I've done something similar myself with Kindle Cloud Reader a few months ago. For the purpose of language learning, I wanted to extract the surrounding context of every word I highlighted in books I read. Turns out it's impossible only with the device.
On Kindle Cloud Reader, Amazon serves images for the pages of the book, so you can't directly read the text. But you can indeed OCR the images and extract the context like that. I created a small browser extension with Claude.
It's crazy how difficult Amazon made this process for reasonable use cases like that.
One thing that has led to me creating way more highlights is KOReader+BookOrbit. Now that I know my highlights aren't locked up in Amazon/Kindle's ecosystem I'm more likely to highlight something.
I had a lot of fun hooking up some e-ink displays in my house to call out to BookOrbit to get the highlights and display them. Highlighting something in my book and then seeing it on a screen a day or two later is more fun than I would have guessed.
I highlighted 1,211 passages in one book. When I exported them from Amazon's notebook page, 283 came back cut off mid-sentence and 180 came back as a location number with no text at all, under a notice that "some highlights have been hidden or truncated due to export limits." Amazon doesn't document the limit and you can't raise it.
Those highlights are my own notes in my own account, so I built a Claude Code skill that gets them back. Three unlocks made it work:
- The Mac Kindle app syncs a SQLite database with the exact character-precise start/end position of every highlight — no export limit applies to it.
- The Cloud Reader renders pages as images that can be captured via canvas and OCR'd locally with Apple Vision (free, on-device, no tokens).
- Known positions turn recovery from transcription into arithmetic: find the known prefix, cut to the known length. Recovered text lands within 0-2 characters of the position ruler (median residual 0-1).
Four books so far: 2,432 highlights, 815 of them export-blocked — all recovered, verbatim, location-cited, into one Markdown file per book.
Constraints, honestly: macOS only (AppleScript browser control, Apple Vision OCR, the Mac Kindle app's data files), your own logged-in Chrome, and Claude Code to drive it. It reads only your own annotations on your own account — no DRM is touched, and the output is for your personal notes.
I've done something similar myself with Kindle Cloud Reader a few months ago. For the purpose of language learning, I wanted to extract the surrounding context of every word I highlighted in books I read. Turns out it's impossible only with the device.
On Kindle Cloud Reader, Amazon serves images for the pages of the book, so you can't directly read the text. But you can indeed OCR the images and extract the context like that. I created a small browser extension with Claude.
It's crazy how difficult Amazon made this process for reasonable use cases like that.
is it impossible now to crack the file directly to get all of the books text out? there used to be a few ways to decrypt them in calibre and similar
Amazon is clamping down on export paths. It's probably only a matter of time before Kindle Cloud Reader gets locked down with DRM too.
Doesn't this skill only work up until Anthropic tweaks weights or the system prompt(s) to not allow it?
One thing that has led to me creating way more highlights is KOReader+BookOrbit. Now that I know my highlights aren't locked up in Amazon/Kindle's ecosystem I'm more likely to highlight something.
I had a lot of fun hooking up some e-ink displays in my house to call out to BookOrbit to get the highlights and display them. Highlighting something in my book and then seeing it on a screen a day or two later is more fun than I would have guessed.
I highlighted 1,211 passages in one book. When I exported them from Amazon's notebook page, 283 came back cut off mid-sentence and 180 came back as a location number with no text at all, under a notice that "some highlights have been hidden or truncated due to export limits." Amazon doesn't document the limit and you can't raise it.
Those highlights are my own notes in my own account, so I built a Claude Code skill that gets them back. Three unlocks made it work:
- The Mac Kindle app syncs a SQLite database with the exact character-precise start/end position of every highlight — no export limit applies to it.
- The Cloud Reader renders pages as images that can be captured via canvas and OCR'd locally with Apple Vision (free, on-device, no tokens).
- Known positions turn recovery from transcription into arithmetic: find the known prefix, cut to the known length. Recovered text lands within 0-2 characters of the position ruler (median residual 0-1).
Four books so far: 2,432 highlights, 815 of them export-blocked — all recovered, verbatim, location-cited, into one Markdown file per book.
Constraints, honestly: macOS only (AppleScript browser control, Apple Vision OCR, the Mac Kindle app's data files), your own logged-in Chrome, and Claude Code to drive it. It reads only your own annotations on your own account — no DRM is touched, and the output is for your personal notes.
Build story: https://baowebdev.substack.com/p/how-to-take-back-your-kindl...
"Don't post generated text or AI-edited text. HN is for conversation between humans."
https://news.ycombinator.com/newsguidelines.html