EPUB to Markdown: A Book That Keeps Its Chapters
Rename an .epub to .zip and unzip it. You'll find a folder of XHTML files, a
stylesheet or two, some images, and a manifest called content.opf that lists every piece and
declares the order they're read in. That's the whole format. An EPUB isn't a mysterious binary blob — it's
a small website in a bag, with a table of contents wired into it.
Which makes it unusually good raw material. When you convert EPUB to Markdown, you're not guessing at structure the way you have to with a PDF. The headings were already headings. Chapter one was already a separate document. Reading order was declared, not inferred. Drop a file into the converter above and you get a Markdown version of the book with its hierarchy intact — free, no sign-up, 50 MB limit, nothing kept on our side.
Why the Structure Survives So Well
Three things inside the container do the heavy lifting, and it's worth knowing what each one contributes:
- The XHTML content files. Real semantic markup —
<h1>,<h2>,<blockquote>,<ol>. These map onto Markdown almost one to one. A chapter heading becomes#, a section becomes##, a pull quote becomes>, italics for a book title stay italics. - The spine. The OPF file lists documents in reading order, so chapters come out
sequenced correctly even when the internal filenames are meaningless things like
part0009.xhtml. - The navigation document. The book's own table of contents. It's the reason chapter titles survive as titles rather than arriving as an anonymous line of bold text.
The practical upshot: the Markdown you get back is navigable. You can scroll to chapter 12, cut it out,
and paste just that into a model. Try doing the same with a 400-page flat text dump and you're using
grep and a lot of hope.
The Bits That Need a Second Look
Nothing here is a dealbreaker, but skimming the output for these takes a minute and saves confusion later.
- Front and back matter comes along for the ride. Copyright page, dedication, "also by this author", acknowledgements, index stubs. In a study workflow that's a few thousand tokens of nothing. Delete it before you paste.
- Footnotes and endnotes get relocated. In the source they're links to a notes file at the end of the book. In flowing Markdown they either cluster at the end of the chapter or appear as bare numbers where the reference was. For fiction, irrelevant. For an academic monograph where half the argument lives in the notes, check where they landed before you build anything on top of it.
- Images don't become text. Cover art, diagrams, and maps are files inside the archive, not words. There's no OCR here, so a diagram's labels won't appear. If a figure carries the explanation, you'll need the original alongside.
- Heading levels can be inconsistent. Some publishers mark every chapter as
<h1>, others use<h1>for parts and<h2>for chapters, and a few use styled<p>tags and no real headings at all. That last case is rare but it's the one that produces a flat-looking conversion from a book that clearly has chapters. - Drop caps and small-caps openings. A chapter starting with a large decorative "I" can surface as a stray character on its own line. Cosmetic, easy to spot.
DRM Stops This Cold — Read This Before Uploading
Most EPUBs bought from a commercial store carry digital rights management. The content files inside the
container are encrypted, and there's an encryption.xml sitting next to the manifest that says
so. No converter can read them, and this one won't either. If you upload a DRM-protected book you'll get
an error or an empty result, and the file is the reason, not the tool.
DRM-free EPUBs are everywhere though. Project Gutenberg, Standard Ebooks, most technical publishers, university presses, Creative Commons releases, anything you exported yourself from a writing tool, and a good chunk of independent stores that sell unencumbered files on principle. Those convert without a fuss.
On the obvious question: converting a book you own so you can study it is ordinary personal use; republishing someone's copyrighted text is a different thing entirely. That's the whole of the advice.
A Whole Book Won't Fit — Work Chapter by Chapter
A full-length novel runs somewhere around 100,000 words, and a dense non-fiction title can be double that. Converted to tokens, that's well past what most models will take in one go, and even the long-context ones get noticeably vaguer as you fill them up. Attention spreads thin. Details in the middle get skipped.
This is exactly where Markdown earns its place over flat text. Because the chapter headings survive, you can slice the file cleanly and work a chapter at a time:
- Summarise each chapter separately, then feed the summaries back in as one short document and ask for the through-line of the whole book.
- Build a study guide per chapter — key claims, terms, questions — and stitch them into one reference.
- Ask targeted questions against a single chapter, so the model's answer comes from 6,000 words rather than being diluted across 150,000.
- Turn the heading tree into an outline first. It's a fast way to decide which three chapters actually matter for what you're doing.
The token counter under the output is the useful signal here. It tells you before you paste whether you're looking at a chapter-sized job or a whole-book one.
EPUB vs the PDF of the Same Book
If you have both, take the EPUB. Every time. The two formats solve opposite problems and it shows in the output.
A PDF is fixed-layout — glyphs pinned to coordinates on a page that never changes size. Extraction means reconstructing sentences from positions, inferring headings from font sizes, and guessing whether two columns should interleave. Running headers and page numbers land in the middle of paragraphs. PDF to Markdown handles all of that well, but it's reconstruction work.
EPUB is reflowable. There are no pages, so there's no page furniture to strip and no column geometry to untangle. The structure is stated rather than implied. The one thing PDF holds onto better is precise typographic layout — complex tables, sidebars positioned just so, anything where the arrangement on the page is part of the meaning. For prose, EPUB wins comfortably.
What People Use This For
- Studying a technical book. Convert an O'Reilly-style title, keep the chapter you're working through, and pair it with your GitHub repository as text so an assistant can see the theory and your actual code together.
- Building a personal reference. The heading hierarchy makes converted books searchable
in Obsidian, a wiki, or a plain folder of
.mdfiles — searchable in a way an e-reader library isn't. - Working through a course reading list. Chapter-level summaries and generated practice questions, traceable back to the section they came from.
- Manuscript work. Authors and editors export an EPUB from Scrivener or Vellum, convert it, and get a diffable, version-controllable copy of the draft.
If you want the whole book as one uninterrupted prose stream instead — for embeddings, readability scoring, or indexing a shelf full of titles at once — EPUB to text is the better fit. The format toggle at the top of this page switches between the two and keeps the file you already selected, so comparing both outputs costs one click rather than a second upload.
Frequently Asked Questions
How do I convert an EPUB to Markdown?
Upload the .epub above and the book comes back as Markdown with its heading structure intact, downloadable as .md. Free, 50 MB per file, no sign-up. EPUB is XHTML underneath, so this is a structure-preserving conversion rather than a lossy extraction — it converts better than almost any other format here.
Why does EPUB convert to Markdown so cleanly?
Because the source is already semantic markup. An EPUB is a zip of XHTML files where a chapter title is genuinely an <h1> and emphasis is genuinely an <em> — the structure is recorded, not implied by how something looks. Mapping that to Markdown is close to a direct translation, which is why headings, lists, blockquotes and italics all survive.
Do chapter headings become Markdown headings?
Yes, at the level the book declared them, so a book with parts and chapters produces a nested outline rather than a flat one. That makes the output genuinely navigable — you can fold it in an editor, link to a section, or ask a model about a specific chapter and have it find the right place.
Are DRM-protected books supported?
No. Encrypted files from Kindle or Apple Books have no readable content until they are decrypted by the reader app that owns them, so conversion returns nothing. DRM-free EPUBs — Gutenberg, Standard Ebooks, most technical publishers, your own manuscripts — work without any special handling.
Is this useful for building a RAG corpus from books?
It is the better starting point of the two formats. Heading structure gives you natural chunk boundaries, so you can split on chapters or sections rather than on an arbitrary character count that cuts mid-argument. If your chunker wants bare prose instead, EPUB to text skips the syntax.
The Rest of the Toolkit
EPUB is one of thirteen formats handled here. File2Txt takes any of them from a single upload if you'd rather not pick a page. For manuscripts and reports there's Word to Markdown, for older rich-text documents RTF to Markdown, and for saved web pages HTML to Markdown. A shelf of books zipped together goes through ZIP to Markdown in one pass.
Working with code or the live web? Convert a GitLab project or a folder on your machine, or point Web2Txt at a URL. The guide to preparing documents for LLMs covers the general shape of all this.
Repo2Txt is built and maintained by v12hero, an independent developer building privacy-first native and web apps.