It was first developed for the Netscape family of browsers starting with Netscape Navigator 2.0 but has subsequently been implemented in other browsers including Mozilla Application Suite, Mozilla Firefox, Safari, Opera, Konqueror, Google Chrome, and some versions of Microsoft Internet Explorer.
Its success can be partly attributed to its simplicity. A plugin declares that it handles certain content types (e.g. "audio/mp3") through exposed file information. When the browser encounters such content type it loads the associated plugin, sets aside the space within the browser content for the plugin to render itself and then streams data to it. The plugin is then responsible for rendering the data as it sees fit, be it visual, audio or otherwise. So a plugin runs in-place within the page, as opposed to older browsers that had to launch an external application to handle unknown content types.
The API requires each plugin to implement and expose a comparatively small number of functions. There are approximately 15 functions in total for initialising, creating, destroying, and positioning plugins. The NPAPI also supports scripting, printing, full screen plugins, windowless plugins and content streaming.
The companies set out the next week to bring what was known as "Allan's Hack" to market. While Netscape was ready to incorporate PDF directly into the browser, and certainly Adobe would have gained from that, Padgett proposed a different approach, a plugin architecture. Adobe engineers Gordon Dow and Nabeel Al-Shamma had recently added a plugin architecture to the Acrobat Reader to leverage the development efforts of engineers outside of the Reader team. Padgett had been a part of that effort, and he expected that if given a chance, other companies (and hopefully teams within Adobe) would choose to extend the Web as well. Clark and team in the end were convinced and set off designing the API that would support the new model. And while PDF was the pioneer, one could argue that later plugins like Macromedia Flash, ActiveX controls, and Java applets have changed the Web landscape far more.
The disadvantage of LiveConnect was that it was tied heavily to the built-in version of Java within the Netscape browser. This prevented the browser from using other Java runtimes, and added a massive amount of bloat to the browser download size since it required Java to script plugins.
Additionally, LiveConnect was tricky to program. The developer had to define a Java class for the plugin, run it through a specialised Java header compiler and implement the native methods. Handling strings, exceptions and other Java objects from C++ was fraught and non-obvious. To compound matters LiveConnect used an earlier and now obsolete API for invoking native C++ calls from Java called JRI. The JRI technology has long been supplanted by JNI.
Mozilla was already using XPCOM to define the interfaces to many objects implemented in C++. Each interface was defined by an IDL (Interface Definition Language) file, and run through an IDL compiler that produced header files and a language neutral type library that was a binary representation of the interface. This binary described the interface, the methods, the parameters, the data structures and enumerations.
This removed the Java dependency, however there are issues with XPConnect. In particular, the technology is heavily based on XPCOM, which is similar to Microsoft COM. Thus the plugin developer must be familiar with reference counting, interfaces, IDL and so forth to implement scripting. Additionally, the dependency on XPCOM led to certain dynamic linking issues (e.g. the fragile base class problem) which had to be solved before the plugin would work correctly with different browsers. XPCOM has since been changed so that it supplies a statically linked version to address such issues. This approach also requires an .xpt file to be installed next to the DLL; otherwise the plugin will appear to work, but scripting won't, causing confusion.
It is supported by the latest generation of Mozilla (1.7.5+) / Firefox, Safari, and Opera. All new plugins should use this API.
A control was implemented as a DLL and loaded into the address space of the host container such as Visual Basic. Earlier versions of Visual Basic used a similar technology called Visual Basic Extensions but OCX controls were seen as superior. Each OCX implemented a well defined subset of the OLE2 interfaces that the container could use to manipulate the control, such as repositioning it, or to provide information about the container. The OCX also implemented an automation interface which allowed it to expose methods and properties that could be manipulated, and used mechanism in the other direction to fire events to the container.
OLE2 was extremely complicated and support for COM in MFC was poor. So Microsoft rationalised the specification to make it simpler, and rebranded the technology as ActiveX. Even after simplification controls were still required to implement about 6 core interfaces. In response to this complexity, Microsoft produced wizards, ATL base classes, macros and C++ language extensions to make it simpler to write controls.
Starting with Internet Explorer 3.0, support was added to host ActiveX controls within HTML content. If the browser encountered a page specifying an ActiveX control via an OBJECT tag (using non-W3C syntax), it would automatically download and install the control with little or no user intervention. This made the web experience "richer" but was perceived as divisive (since controls only ran on Windows) and a security risk due to the lack of user intervention. Microsoft has been forced to introduce security measures to address its shortcomings. For example:
Internet Explorer did for a time support NPAPI plugins. Plugins that functioned in the Netscape browser also functioned in Internet Explorer. This was due to a small ActiveX control implemented within a "plugin.ocx" file that acted as a shim between the ActiveX based browser and the NPAPI plugin. The IE browser would load the control and use it to host plugins specified within the page. However, Microsoft made the claim that the NPAPI plugins (or the IE implementation of the API) were a security issue and dropped support for them in version 5.5 SP2.
One important difference between NPAPI and ActiveX is that NPAPI is solely for Internet plugins, while ActiveX is used for a wide variety of purposes, including application composition in Visual Basic. A typical Windows user has a vast array of ActiveX controls installed, a number of which are probably marked "safe for scripting", but are not actually secure. Any of these can be used as angles to subvert the user's computer.
Another difference for the NPAPI is that implementations (prior to Mozilla Firefox, see below) did not automatically download or install missing plugins. A missing plugin caused the browser to display a jigsaw piece representing the plugin. If the user clicked on that they were directed to Netscape's plugin finder service where they could manually download and install the plugin for themselves. While this is inconvenient to the user, it is also an important security measure since it prevented the content using the browser as a vector for malware.
In Internet Explorer, the HTML content specifies the location where the ActiveX control resides. If the control is not already installed, IE will automatically download and install the control from the specified source, pausing only to show the digital signature to the user and obtain their consent for installation to start. For legitimate controls, this offers a more streamlined installation mechanism with minimal user interaction. However malicious content could convince the user with clever social engineering to ignore warnings (or their better judgement) and install something that might harm their privacy or the machine. A number of spyware, adware and malware sites use this mechanism to deploy executable content to machines. Microsoft has had to increase the default security settings for ActiveX and maintain blacklists of malicious controls in an attempt to mitigate this risk.
Mozilla Firefox attempts to present a middle ground. If a plugin is missing, it will notify the user that the plugin is missing and initiate a secure connection to a plugin finder service hosted on mozilla.org. The user can permit Firefox to download and install the plugin. This model prevents content specifying where a plugin should be downloaded from – the plugin finder service does. This enables Firefox to present a fairly seamless installation mechanism but limit the service to trusted and compatible plugins from reliable sources. Of course this model implicitly trusts the plugin finder service to return "good" plugins, increasing the security required on the host site.