The underlying annotators were trained using a variant of the Yarowsky algorithm and therefore take context into account for things like 差 (chāi or chà) and 为 (wèi or wéi) although it still makes some mistakes. Training data included proprietary documents, so, although the apps are free, the data cannot be fully open-source, and annoyingly CC-CEDICT could not be used because its CC license conflicted with this situation. But there are some limited English definitions available, or the apps will link to Pleco or Hanping if you have either installed (tap on any word to access the popup definitions and Pleco/Hanping button, unless that word is part of a web link, in which case tapping on it just opens the link as normal). The code without the data is Apache-licensed if you want to re-train it on another corpus yourself, but the results are not likely to be as good unless you have access to a better corpus than I do.
There exists code for an iOS version, but the Apple developer programme is incredibly unhelpful (1 they only accept apps compiled on a decent Mac with latest OS, not a few-years-old one, 2 they demand 100 pounds a year to subscribe to the programme, universities can be exempt but individual university members cannot, and 3 I have no idea how much extra they invoice for testing new releases). There are however a couple of established iOS developers who have been given the back-end code, so let's hope they release browsers soon.
As for desktop browsers, I have had some success making "bookmarklets" (like mini browser extensions), but these rely on sending the page to a third-party server to get the pinyin and are therefore not great for privacy unless you can also set up the annotation server on your own machine. Also an increasing number of sites are blocking the use of bookmarklets altogether in their "Content Security Policy" (CSP), so if you want bookmarklets you now need a browser extension to turn off CSP as well (or a proxy that does the job, but then the proxy has to rewrite secure HTTPS into insecure HTTP; an increasing number of sites have user-side scripts that break when this happens). There is however a Windows 10 developer who has the back-end code and is making a Win10 "universal app" for both desktops and mobile devices. In all cases the problem remains of fixing up the sites that are too fussy about layout. (You could work around this by cramming pinyin into the very small existing space between the lines, but the result would not be half as easy to read.)