MOBI to EPUB Converter

A MOBI file is a Palm Database (PDB) binary container following the PalmOS database specification. It opens with a 78-byte header: a database name (32 bytes), attributes, version, creation/modification times, a type identifier (usually BOOKMOBI), a creator identifier (MOBI or TEXt), the record count, and a record offset table - one entry per record giving its byte position in the file. Everything (text, images, indices) is chopped into roughly equal chunks of about 4096 bytes stored sequentially as records: record 0 holds the headers, records 1 through N the compressed text, later records the images.

Record 0 packs three consecutive headers. The PalmDOC header declares the compression scheme (1 = none, 2 = PalmDOC LZ77 dictionary compression, other values meaning Huffdic Huffman dictionaries), total text length, text record count, per-record size, and encryption type - a nonzero encryption field marks DRM, which the converter refuses to touch. The MOBI header adds text encoding (CP1252 or UTF-8), the index of the first image record (firstImageIndex), and the EXTH flag. When set, the EXTH header follows immediately: a series of type-length-value entries carrying author, updated title, publisher, ISBN/ASIN, publishing date and similar metadata.

The body text decompresses record-by-record into a restricted HTML dialect close to HTML 3.2/4: limited CSS support, no SVG or embedded fonts, images referenced as <img recindex="00001"> pointing at the nth image record after firstImageIndex, and table-of-contents anchors written as byte offsets like <a filepos="0000023289">. To become EPUB this markup must be upgraded to well-formed XHTML: tags closed and nested correctly, attributes quoted, unsupported constructs folded into safe equivalents, recindex references rewritten into manifest image entries such as <img src="images/xxx.jpg">, and every filepos target given an explicit id attribute so chapter links keep working.

Assembling the EPUB then means building the OCF ZIP container. The mimetype entry (containing exactly application/epub+zip) must be the first file in the archive and stored uncompressed; META-INF/container.xml points to the package document; content.opf lists every component in the manifest and fixes the linear reading order in the spine, while its metadata block carries Dublin Core terms mapped from EXTH fields (author to dc:creator, updated title to dc:title, and so on). Navigation is written twice - an EPUB 3 nav.xhtml for modern readers and a legacy toc.ncx for backward compatibility. On the desktop side, Calibre's ebook-convert ships a mature MOBI input plugin and remains the de facto command-line standard, and KindleUnpack can split a MOBI into HTML plus images beforehand. This tool only processes DRM-free MOBI - Amazon-encrypted purchases cannot be parsed - and .azw3/KF8 files use a different, EPUB-like structure requiring their own converter.

• PDB container structure: 78-byte Palm database header (database name, type BOOKMOBI, creator, record offset table); all content split into roughly 4096-byte records stored sequentially
• Record 0 holds three headers: PalmDOC header (compression scheme, text length, encryption type) -> MOBI header (encoding CP1252/UTF-8, firstImageIndex, EXTH flag) -> EXTH header (type-length-value metadata such as author and title)
• Text recovery: records 1 through N are decompressed per the PalmDOC header (none / PalmDOC LZ77 / Huffdic Huffman) and concatenated into the source HTML
• HTML upgrade: the restricted HTML is normalized to well-formed XHTML; recindex image references become manifest files; filepos byte anchors become explicit id targets
• Metadata mapping: EXTH fields map onto Dublin Core terms (dc:title, dc:creator, dc:language...) in the OPF metadata block
• OCF packaging: uncompressed mimetype entry first, META-INF/container.xml, content.opf manifest + spine reading order, navigation written both as EPUB 3 nav.xhtml and legacy toc.ncx
• Limits & ecosystem: DRM-free input only; Calibre ebook-convert and KindleUnpack cover desktop needs; Amazon's Send to Kindle accepts only EPUB since late 2022