Accessible PDF View

A browser extension that rebuilds PDFs as structured HTML. Headings, lists, tables — with row and column headers — and links come out as real HTML elements, so a screen reader user can read a PDF the way they read any other web page: by heading, by list, by table navigation. ■ Three views Reader — the document rebuilt as HTML. Where a PDF carries structure tags, the author's own structure is used; where it does not, structure is inferred from the layout. For a tagged PDF you can switch between the two and compare. Original — the original PDF pages, which work even when no text can be extracted. Markdown — copyable text: either what the Reader shows, or the parser's raw output. ■ It does not assert what you cannot check ・An image with no description is reported as having none. ・Alternative text written by the author, generated by the authoring tool, and generated by this extension are told apart, and the source is always shown. ・Machine-read text is labelled as machine-read every time it is read. ・When a page yields no text, the extension finds out whether it is an image, text drawn as outlines, or genuinely empty — and says which. ■ Settings Font size, line height, column width, typeface and theme, from the browser's side panel. Font size is a multiple of your browser's own default rather than a fixed size, so it respects the setting you already made. Forced-colors mode is supported. Atkinson Hyperlegible, designed for low vision, is packaged with the extension, so choosing it needs no installation — Latin script only, with Japanese drawn by the system face beside it. Also: the language of the interface, the language AI-generated text is written in, which reading a tagged PDF opens with, and which viewer opens a PDF. ■ Opening PDFs Normally you open a PDF and press the toolbar button. Switch the setting and addresses ending in .pdf open in the Reader directly instead; the browser asks for access at the moment you switch it on, and gets it back when you switch it off. Leave the setting alone and nothing is ever asked for. ■ Privacy PDFs are parsed inside the browser. In ordinary use no document is sent to any server, and no usage data is collected. Image descriptions and OCR are optional and use the browser's own on-device AI model where one is available. No image leaves the machine. It asks for five permissions at install — activeTab, storage, contextMenus, sidePanel and declarativeNetRequestWithHostAccess — and none of them is a host permission: installing it grants access to no site. One optional host permission exists, and nothing asks for it until you do something that needs it: one site, when a PDF could not be read from it, or every site, for the Opening PDFs setting. Both are revocable. ■ Known limits ・Local files (file://) cannot be opened. ・Image descriptions and OCR use Chrome's built-in AI and are available in Chrome only. On Firefox those panels leave themselves out and say why. ・For PDFs without structure tags, headings and tables are inferred and can differ from the real structure. The Reader says when inference is in use. ・OCR is experimental, and the on-device model has real accuracy limits. Open source, MIT licensed. https://github.com/motchie/accessible-pdf-view https://accessiblepdfview.org