Installation

The recommended way is not to install at all and run it through uvx as shown above. uv itself is a single binary; see its install page.

If you would rather have a permanent ebook-convert command on your $PATH:

$ uv tool install ebook-convert      # with uv
$ pipx install ebook-convert         # with pipx
$ pip install ebook-convert          # into the current environment

Single-file executables for Linux (x86_64, aarch64) and macOS (arm64, x86_64) that need no Python at all are attached to each GitHub release.

Optional pieces:

• pdftohtml, pdfinfo and pdftoppm from poppler are needed in $PATH for conversion from PDF.

• pip install 'ebook-convert[markdown]' enables Markdown input (.md); [japanese] adds Japanese transliteration for --asciiize.

• optipng, jpegtran/cjpeg and cwebp are used to optimise images when present; Pillow is used otherwise.

Usage

$ ebook-convert INPUT OUTPUT [options]

The command line is Calibre’s own ebook-convert command, unchanged: the output format is taken from the extension of OUTPUT (.EXT alone derives the name from the input, no extension writes an OEB folder), and all of Calibre’s conversion options work, from --authors and --cover to --base-font-size, --chapter, --level1-toc, --extra-css, --embed-font-family, --search-replace and the heuristic processing switches. ebook-convert in.epub out.mobi -h lists the options that apply to that pair of formats. The Calibre conversion manual applies verbatim.

Diagnostics go to stderr; -v makes them more verbose, --debug-pipeline DIR dumps the intermediate stages.

Supported formats

Every entry below is exercised by tools/survey.py with a real conversion (output: EPUB fixture to the format; input: that result back to EPUB, text checked to survive).

Input: azw, azw3, azw4, docx, docm, epub, fb2, fbz, htm, html, xhtml, shtml, htmlz, lrf, md/markdown (needs the markdown extra), mobi, prc, pobi, odt, opf, pdb, updb, pdf (needs poppler), pml, pmlz, rb, rtf, snb, tcr, txt, text, textile, txtz, zip (of HTML), kepub (also as .kepub.epub; the Kobo markup is stripped).

Output: azw3, docx, epub, fb2, htmlz, kepub (or .kepub.epub, Kobo’s own naming), lrf, mobi, oeb (a folder), pdb, pmlz, rb, rtf, snb, tcr, txt, txtz, zip (Calibre’s “HTML output”: a zip of HTML files with an index).

Compared with the 4.x line this adds HTML/zip, PDB, PMLZ, RB, RTF, SNB and TCR output, KEPUB input and output, and PML, RB, SNB, TCR and AZW4 input, plus everything Calibre changed in the converters between 2020 and 2026 (EPUB 3 navigation, KF8 fixes, DOCX tables and footnotes, the new CSS parser, …).

Not included, and why:

• PDF output: rendered by Calibre with Qt WebEngine and PoDoFo (C++).

• LIT input/output: needs the lzx/msdes C extensions.

• CHM and DjVu input: need the chm and bzzdec C extensions.

• Comic (CBZ/CBR) input: image processing with Qt, unrar.

• News download (recipes): needs Calibre’s web framework.

• RAR archives anywhere: need unrardll.

• Metadata download, library, devices, viewer, editor, server, AI features: not conversion.

Differences from Calibre

A handful of Calibre’s building blocks are C/C++ or Qt and have been replaced by pure-Python equivalents. These are the places where behaviour can differ from running ebook-convert inside a Calibre install:

• Sentence splitting (calibre.spell.break_iterator): the Kobo <span> per sentence in KEPUB output is placed by ICU’s sentence iterator in Calibre. Here it follows kepubify’s rule instead (a sentence ends at ., ! or ?, an optional closing quote or ellipsis, and whitespace), which is what Kobo’s own tooling produces.

• ICU (calibre.utils.icu): sorting, case changes and case-insensitive searches are done with Unicode normalisation and case folding from the standard library. Results are locale independent (no Swedish v/w merging, no Turkish i rules, no language specific contractions), and bidi reordering of PDF text (visual_to_logical) is an identity transform. Sorting is only used for things like index pages and manifest ordering; for conversion output the difference is cosmetic.

• Images (calibre.utils.img): re-implemented on Pillow. Covers, image rescaling, grayscale/dither for e-ink profiles, GIF/PNG/JPEG/WebP conversion all work; the quality of resampling and JPEG encoding is Pillow’s rather than Qt’s. JPEG-XR needs the external JxrDecApp.

• SVG rasterisation: rendered by resvg (resvg_py wheel) instead of QSvgRenderer. resvg is a complete SVG 1.1 renderer; SVG fonts and CSS @font-face inside SVG are not supported.

• Generated covers (the “default cover” EPUB output adds when a book has none): a plain title/series/author cover drawn with Pillow, instead of Calibre’s themed, template-driven Qt painter.

• HTML parsing: html5lib instead of the html5-parser C extension. Same HTML5 algorithm, slower, and element line numbers are not recorded.

• Encoding detection: chardet instead of uchardet.

• PalmDOC compression (MOBI/PDB): the Python implementation; slower on very large books.

• Translations: none, _() is the identity. All messages are English.

• Fonts: no FreeType/fontconfig; the system font scanner reads font files directly (--embed-font-family works with fonts from ~/.fonts, /usr/share/fonts etc.). Liberation fonts are not bundled, so LRF output and font embedding fall back to whatever the system has.

• Configuration lives in ~/.config/ebook-convert (XDG), not in Calibre’s directory; Calibre tweaks can be put in a tweaks.json there. Set EBOOK_CONVERT_CONFIG_DIRECTORY to move it.

• Workers: jobs Calibre runs in a helper process (pdfinfo metadata, exploding AZW3 files) run in-process.

• Reading PDF metadata works (pdfinfo); writing PDF metadata does not (PoDoFo).

How this package is built

ebook_convert/ is generated; do not edit it by hand. Everything that turns a Calibre checkout into this package lives in tools/:

tools/vendor.py (standard library only)

python tools/vendor.py /path/to/calibre rebuilds ebook_convert/ from scratch, deterministically:

1. copies an explicit include list (src/calibre/ebooks, customize, the needed utils modules, polyglot, tinycss, css_selectors and the resource files under resources/ into ebook_convert/data/);

2. deletes an explicit exclude list (LIT/CHM/DjVu/comic/PDF-output/recipe plugins, editor-only oeb/polish modules, metadata download sources, C sources, tests) and every non-Python file outside data/;

3. rewrites imports: calibre to ebook_convert, polyglot, tinycss, css_selectors, html5_parser and calibre_extensions to the copies/stand-ins inside ebook_convert;

4. makes Calibre’s Python 3.14 code run on 3.12 and 3.13 (except A, B: to except (A, B):, from compression import zlib to import zlib, type parameter defaults dropped, from __future__ import annotations added for the deferred annotations Calibre relies on);

5. applies the ordered patch rules in tools/patches/ (small exact text replacements; a rule that no longer matches makes the run fail, which is the point: after a Calibre upgrade you fix or drop the rule);

6. copies the hand-written modules in tools/overlay/ over the tree;

7. writes ebook_convert/_vendored.py with the Calibre commit.

python tools/vendor.py --check lists imports of calibre, qt, PyQt or calibre_extensions still present; module-level ones fail the check, indented (lazy) ones are listed for information.

tools/patches/

One file per patch, applied in name order:

• 010_localization_iso639: build the ISO-639 language tables from the bundled iso_639-3.json instead of Calibre’s build-time msgpack file.

• 020_customize_builtins: keep only the archive, metadata and conversion plugins and the profiles; register the verified plugin set.

• 030_customize_ui: placeholder classes for device drivers and metadata sources in the plugin registry.

• 040_formatter_functions: inline two constants from calibre.db.

• 050_conversion_cli: no recipe listing; log to stderr; usage text.

• 060_docx_paper_sizes: inline the paper size table from the PDF renderer.

• 070_lrf_device_stub: inline DevicePlugin.THUMBNAIL_HEIGHT.

• 080_metadata_pdf_podofo: PDF metadata writing raises instead of importing PoDoFo.

• 090_txt_input_txtz_metadata: read TXTZ metadata from the archive, not from the inner .txt (fixes lost title/author).

• 100_markdown_series: inline the “Series [index]” parser from calibre.db.

tools/overlay/

Replacements and additions, kept small and documented in each file:

• constants.py, startup.py, utils/resources.py: trimmed constants, process initialisation (_, P(), I() builtins, mimetypes, temp dir) and resource lookup inside ebook_convert/data.

• utils/icu.py: the pure-Python ICU replacement described above.

• utils/img.py: Pillow image utilities (from the 4.x line, plus WebP and alt-text helpers from current Calibre).

• utils/html5.py, html5_parser/: html5lib-based html5_parser.

• calibre_extensions/: Python stand-ins for the C extensions speedup, fast_html_entities, cPalmdoc, uchardet, translator and tokenizer (CSS Color 4 parser); any other extension raises ImportError so Calibre’s own fallbacks kick in.

• utils/ipc/simple_worker.py, utils/safe_atexit.py: in-process replacements for Calibre’s worker processes.

• ebooks/oeb/transforms/rasterize.py: resvg rasteriser.

• ebooks/covers.py: Pillow cover generator.

• ebooks/metadata/sources/identify.py: urls_from_identifiers only.

• main.py: the console entry point.

• data/iso_639-3.json: language codes (from the iso-codes project).

tools/check_imports.py

Imports every module under ebook_convert/ and lists failures.

tools/survey.py

Runs a real conversion through every registered input and output plugin and prints a table.

$ git clone --depth 1 https://github.com/kovidgoyal/calibre.git /tmp/calibre
$ ./make.sh vendor /tmp/calibre      # vendor + --check + import check
$ ./make.sh test
$ ./make.sh survey

Development

$ ./make.sh install     # uv venv with all extras
$ ./make.sh test        # pytest (the 4.x acceptance suite + tree checks)
$ ./make.sh check       # vendor --check + import every module
$ ./make.sh survey      # convert through every plugin
$ ./make.sh build       # sdist + wheel + twine check
$ ./make.sh binary      # PyInstaller single-file executable

CI runs the test-suite on Python 3.12, 3.13 and 3.14 on Linux and on macOS, builds the wheel and smoke-tests it with uvx.

Related projects

Tools that get books into a format this converter understands, or that cover ground it does not:

• FicHub: paste a fanfiction URL, get an EPUB (also MOBI, PDF or zipped HTML). Covers FanFiction.net, FictionPress, SpaceBattles, SufficientVelocity, QuestionableQuesting and more. Source is on GitHub.

• WebToEpub: browser extension (Chrome, Firefox) that turns the web novel or serial you are reading into an EPUB for offline reading; supports Archive of Our Own, FanFiction.net, Royal Road, Baka-Tsuki and many others. Code is on GitHub.

• FanFicFare: command-line tool and Calibre plugin that downloads fanfiction from a long list of sites into EPUB, with metadata.

• Calibre: the real thing, with GUI, library, viewer, editor and every format. Use it if you can install it.

• pandoc: converts between document formats including EPUB and DOCX, from Markdown and LaTeX; no MOBI/AZW3/LRF.

• kepubify: EPUB to Kobo KEPUB, in Go; the sentence rule for Kobo spans here is taken from it.

• KCC: comics to e-reader formats.

License

GPL-3.0-or-later, like Calibre. Copyright Kovid Goyal and the Calibre contributors for the converted code (headers are kept in every file); the tooling and overlay modules are copyright their respective authors as noted in the files.