Download websites that use MathJax

About MathJax

Very few web technologies tell you what a page is about. MathJax does. A site loads the engine because its content contains mathematics that has to render as mathematics rather than as an image, which makes the fingerprint a subject classifier wearing the clothes of a JavaScript dependency.

The provenance supports that reading. The engine dates from 2009, is written in JavaScript, is published under Apache 2.0 and is maintained by the MathJax Consortium with backing from scholarly and educational organisations. Nothing about it is a growth product. Adoption came from journals, university departments, courseware, technical documentation, engineering and quantitative teams publishing formulas, and reference sites that needed notation to survive being copied into another page.

As of September 2026 the renderer is detected on 27,618 sites, 15,451 of them in United States, with domain volume weighted toward .com. A generic domain suffix is a weak proxy for institution type here, because research groups, publishers and commercial technical sites all register under the same one. Filtering the download by an academic or government suffix isolates institutions far more reliably than reading the headline geography does.

Version data adds a second dimension. We record 40 distinct version strings, with readings clustering on 2.7.9, a build from the earlier generation of the engine rather than the rewritten line that replaced it. A distribution shaped that way is a maintenance map. Mathematical rendering is set up once, usually by whoever built the site, and then left alone, so the version string sits close to a timestamp of the last time anybody touched the front end. Pages still on the older line load a different configuration format and behave differently on accessibility and performance, which matters to anyone selling into that stack or auditing it.

Publishers, education technology vendors, accessibility auditors and performance tooling companies all want the same cut, and none of them can assemble it one domain at a time, because the underlying question is which sites publish mathematics at all rather than what any single site runs.

Pairing the field with the rest of a domain's stack is where the cut earns its keep. An institutional repository platform, a citation tool, a course delivery system or a preprint server beside the renderer identifies academic publishing without any need to read the domain suffix. A documentation generator beside it identifies a software project explaining an algorithm. A commercial content system beside it identifies a company explaining a model to its customers. The renderer says mathematics is on the page, and the rest of the stack says who put it there and why. Watching the population move between crawls adds a third reading, because a subject specific dependency shrinks when its field adopts a different tool rather than when the web at large changes direction.

The caveat is the rendering path. Detection depends on the engine loading in the browser, whether from a public content delivery network or a local copy, so pages that pre render their notation on the server or ship static images of formulas leave no trace. Technical publishing has been drifting quietly in that direction. Read the count as the population rendering mathematics in the browser today, and expect the shape of that population, rather than its total, to be the durable part.