A Chrome extension that exports pages already rendered by Amazon Kindle Cloud Reader to a PDF on a real paper size, optionally with an invisible OCR text layer so the pages are searchable and selectable.
Use it only to make personal copies of books you own and are authorized to copy.
Everything runs locally in your browser. Captured pages, book identifiers, and OCR data are never uploaded or sent to an external API or cloud AI service. It does not download Kindle book files, decrypt content, bypass DRM, or send telemetry. The OCR engine and English model are bundled with the extension.
No git, npm or build needed.
- Download the
.zipfrom the latest release. - Unzip it, and move the unzipped folder somewhere you will keep it. Chrome loads the extension from that folder every time it starts, so deleting it uninstalls the extension.
- Open
chrome://extensionsand turn on Developer mode, top right. - Click Load unpacked and select the unzipped folder itself.
- Open a book at https://read.amazon.com/ (regional domains such as
read.amazon.co.ukandread.amazon.com.brwork too) and click the extension.
Developer mode has to stay on: Chrome installs packaged extensions from the Chrome Web Store and nowhere else, and this is not published there on purpose. Updating means downloading a newer zip and repeating step 4.
Cloning the repository and running Load unpacked on the clone works exactly the same way. The bundles and the OCR engine are checked in. This is the way to go if you intend to change anything.
Click the extension and choose Scan / resume. The same button says Stop while a scan is running. Keep the reader tab visible while it scans, and if pages come out skipped, choose the slower page delay.
The scan stops on its own at the end of the book, or when the reader stops advancing. Captured pages are cached per book, so you can stop whenever you like and pick up later — a page already captured is never re-encoded, and an identical view is never stored twice.
Everything that acts on the whole book lives behind See captured pages and save PDF: reordering, deleting single pages, Page size, Add searchable text (OCR), the two save buttons, and Clear this book.
Page size writes every page onto one real sheet — A4, US Letter, or Match capture to keep whatever pixel size the reader happened to paint. The bitmap is fitted and centred, so a page that does not match the sheet gets even bands rather than being cropped or stretched.
In two-column mode the reader paints a whole spread as one landscape image. Split two-page spreads (on by default) cuts those down the middle into two portrait pages. Portrait captures are never touched, so switching the reader to single column turns splitting off by itself.
The reader paints pages as images, so there is no text to copy — the only route to a text layer is to read the pixels back. Add searchable text (OCR) runs Tesseract over the cached pages and stores the words with their positions. Reckon on a second or two a page, so five to ten minutes for a long book. It runs in the preview tab, which has to stay open but does not have to be the visible one, and it is resumable: results are saved as they finish, and closing the tab costs at most the pages in flight.
Accuracy comes from the pixels, so it is only as good as the capture. On clean printed prose the Balanced JPEG quality is enough; High is the first thing to change if a book comes out badly. The engine is English only.
You do not need to re-scan a book captured before you decided you wanted text.
Save PDF is the scan: one image per page, with an invisible text layer over any page OCR has read. It looks exactly like the book, whether or not you ran OCR at all.
Save text PDF drops the bitmaps and sets the recognised words as real type, placed where they were found, so headings, indents and margins keep their places. The file comes out on the order of a hundredth the size and stays sharp at any zoom. What it costs is everything that was not text — rules, ornaments and drop caps do not survive, and a misread word is a misread word rather than a picture of the right one. It needs OCR to have run first.
Plates are the exception. Every tile in the preview grid has an img box: tick it and that page goes into the text PDF as the image, with the invisible words still over the top. Pages that are mostly picture get ticked for you, as do pages OCR failed on, which are exactly the ones to keep as bitmaps. Once you touch a box, your answer stands and no later run overrules it.
Expect it to. Kindle Cloud Reader is not a stable target and Amazon has no reason to make it one: the class names, the footer counter, the key handling and the page markup are all theirs to change, and a rendering path that quietly ignores a synthetic key event costs them nothing to ship. Assume any break lands in the direction of making it harder to keep the books you paid for, not easier, and assume no upstream fix arrives on your schedule.
So fork it. This repo is meant to be a foundation rather than a product — the tedious parts are the ones that are done, and they are the ones that do not depend on Amazon's markup: the per-book cache and resume, the PDF writer that streams a long book without holding it in memory, the invisible text layer, the paper fitting, the unit tests. What breaks is the thin layer that reads the page and turns it, and that is the part a coding agent can re-fit quickly if you hand it the symptom and let it read the rest.
Start in src/content/. Every assumption about Amazon's markup is in dom.ts,
which is the whole blast radius of a reader change. A few decisions in there look
redundant and are not — trusting the page counter instead of hashing pixels, or
turning pages the obvious way — and each one is commented where it is made, with
the failure it avoids. Read those before letting an agent tidy them: they all fail
silently, by dropping pages from a scan that otherwise looks like it worked.
npm run fixture serves two offline pages that need no account and no book:
test/fixture.html has real reader markup with synthetic pages, and
test/ocr-fixture.html runs the OCR engine over known text. They are the fastest
way for you or an agent to tell a broken change from a broken capture.
Only needed if you change src/.
npm install
npm run build # typecheck, then bundle src/ to content.js, popup.js, preview.js
npm run check # typecheck, lint and testsnpm run watch rebuilds on change. npm run pack writes the release zip to
dist/ — the files Chrome loads and nothing else; pushing a v<version> tag
matching manifest.json makes GitHub Actions build that zip and publish it as a
release. npm run vendor re-copies the OCR engine from
node_modules into vendor/tesseract; that directory is checked in so Load
unpacked works on a fresh clone, and its licenses and upstream versions are
listed in vendor/tesseract/THIRD_PARTY_NOTICES.md.
What the extension has access to is a rendered page in your own browser, in a book your own account has already opened. It reads pixels off the screen — the same pixels you are looking at — and writes them to a file.
It does not do any of the following, and there is no code here that could:
- download, request or open a Kindle book file of any kind
- decrypt anything, extract or derive a key, or interact with DRM at all
- bypass a paywall, a licence check, a lending expiry or a device limit
- give access to a book the signed-in account cannot already read
- send a page, a book identifier or anything else anywhere; the OCR engine is bundled rather than fetched, and there is no other network code
So it is a screen capture tool with a page-turner and a PDF writer attached, not a DRM tool. That distinction is the whole design, not a disclaimer.
None of that decides whether you may copy a particular book. Copyright in the text is untouched by any of this, and capturing a book you do not have the right to copy is on you. Automated capture may also conflict with Amazon's Conditions of Use regardless of copyright; that is between you and Amazon, and using this may put your account at risk. Nothing here is legal advice.
Not affiliated with, authorized by, or endorsed by Amazon. "Amazon", "Kindle" and "Kindle Cloud Reader" are trademarks of Amazon.com, Inc. or its affiliates, used here only to say which website this software works with.
Bug fixes are welcome. New features are out of scope for this repo — if you want it to do something else, fork it; that is what it is for. See CONTRIBUTING.md.
MIT © Nikolai Andreadi.
The Tesseract OCR engine and English model in vendor/tesseract are third-party
code under the Apache License 2.0 and keep their own terms.