markmatsu/pdf-batch-translator

This is a command line tool that takes a folder full of English PDFs and translates each one into Japanese, saving the result as a Markdown file. It is built to be more careful than a simple loop that just feeds each page to a translation API. First, it tries to translate only the actual body text of a document. Things like navigation menus, search boxes, ads, page numbers, headers and footers, footnotes, bibliographies, and image credits get stripped out before translation, so the reader ends up with a clean version of just the real content. Lines that repeat across many pages are removed automatically, and the tool relies on the AI model to make the final call on what counts as body text versus clutter. Second, the translation is meant to be literal rather than loosely paraphrased, so word order, paragraphs, and headings stay close to the original structure in the final Markdown output. Third, the tool carries a bit of context from the end of the previous translated section into each new request, so terminology and sentence flow stay consistent across chapters instead of every chunk being translated as if it exists in isolation. The tool writes its output to disk after every chunk is translated, so if you interrupt it midway, the work done so far is saved safely rather than lost or left in a broken state. It also skips PDFs that already have an output file, so re-running it on a folder only processes the files you have not translated yet, and by default it caps how many chunks of a very long PDF it will translate in one run as a safety measure against runaway usage. To use it, you need Python, an Anthropic API key, and to install a short list of dependencies including PyMuPDF for reading PDFs. You can run it on either a whole folder or a single file, and options let you change the AI model used, adjust chunk sizes, supply a glossary of preferred translations, and control the safety cap on chunks.