17 KiB
| label | order | icon |
|---|---|---|
| Creating your own providers | 998 | repo |
Creating your own Providers
Providers in CloudStream consists primarily of 4 different parts:
When making a provider it is important that you are confident you can scrape the video links first! Video links are often the most protected part of the website and if you cannot scrape them then the provider is useless.
0. Scraping
If you are unfamiliar with the concept of scraping, you should probably start by reading this guide which should hopefuly familiarize you with this technique.
Looking at how some extensions work alongside reading this will likely help a lot. See what common patterns you can spot in multiple extensions.
1. Searching
This one is probably the easiest, based on a query you should return a list of SearchResponse
Scraping the search results is essentially just finding the search item elements on the site (red box) and looking in them to find name, url and poster url and put the data in a SearchResponse.
The code for the search then ideally looks something like
// (code for Eja.tv)
override suspend fun search(query: String): List<SearchResponse> {
return app.post(
mainUrl, data = mapOf("search" to query) // Fetch the search data
).document // Convert the response to a searchable document
.select("div.card-body") // Only select the search items using a CSS selector
.mapNotNull { // Convert all html elements to SearchResponses and filter out the null search results
it.toSearchResponse()
}
}
// Converts a html element to a useable search response
// Basically just look in the element for sub-elements with the data you want
private fun Element.toSearchResponse(): LiveSearchResponse? {
// If no link element then it's no a valid search response
val link = this.select("div.alternative a").last() ?: return null
// fixUrl is a built in function to convert urls like /watch?v=..... to https://www.youtube.com/watch?v=.....
val href = fixUrl(link.attr("href"))
val img = this.selectFirst("div.thumb img")
// Optional parameter, scraping languages are not required but might be nice on some sites
val lang = this.selectFirst(".card-title > a")?.attr("href")?.removePrefix("?country=")
?.replace("int", "eu") //international -> European Union 🇪🇺
// There are many types of searchresponses but mostly you will be using AnimeSearchResponse, MovieSearchResponse
// and TvSeriesSearchResponse, all with different parameters (like episode count)
return LiveSearchResponse(
// Kinda hack way to get the title
img?.attr("alt")?.replaceFirst("Watch ", "") ?: return null,
href,
this@EjaTv.name,
TvType.Live,
fixUrl(img.attr("src")),
lang = lang
)
}
In this code snippet I have separated the Element to SearchResult conversion to a separate function because that function can often be used when scraping the home page later. No need to parse the same type of element twice.
2. Loading the home page
Getting the homepage is essentially the same as getting search results but with a twist: you define the queries in a variable like this:
override val mainPage = mainPageOf(
Pair("1", "Recent Release - Sub"),
Pair("2", "Recent Release - Dub"),
Pair("3", "Recent Release - Chinese"),
)
This dictates what the getMainPage function will be receiving as function arguments. Basically when the recent dubbed shows should be loaded the getMainPage gets called with a page number and the request you defined above.
override suspend fun getMainPage(
page: Int,
request : MainPageRequest
): HomePageResponse {
// page: An integer > 0, starts on 1 and counts up, Depends on how much the user has scrolled.
// request.data == "2"
// request.name == "Recent Release - Dub"
With these variables you should fetch the appropriate list of Search Response like this:
// Gogoanime
override suspend fun getMainPage(
page: Int,
request : MainPageRequest
): HomePageResponse {
// Use the data you defined earlier in the pair to send the request you want.
val params = mapOf("page" to page.toString(), "type" to request.data)
val html = app.get(
"https://ajax.gogo-load.com/ajax/page-recent-release.html",
headers = headers,
params = params
)
val isSub = listOf(1, 3).contains(request.data.toInt())
// In this case a regex is used to get all the correct variables
// But if you defined the Element.toSearchResponse() earlier you can often times use it on the homepage
val home = parseRegex.findAll(html.text).map {
val (link, epNum, title, poster) = it.destructured
newAnimeSearchResponse(title, link) {
this.posterUrl = poster
addDubStatus(!isSub, epNum.toIntOrNull())
}
}.toList()
// Return a list of search responses mapped to the request name defined earlier.
return newHomePageResponse(request.name, home)
}
This might seem needlessly convoluted, but this system is to allow "infinite" loading, e.g loading the next page of search responses when the user has scrolled to the end.
TLDR: Exactly like searching but you defined your own queries.
3. Loading the show page
The show page is a bit more complex than search results, but it uses the same logic used to get search results: using CSS selectors and regex to parse html into a kotlin object. With the amount of info being parsed this function can get quite big, but the fundamentals are still pretty simple. The only difficultuy is getting the episodes, they are not always not part of the html. Check if any extra requests are sent in your browser when visiting the episodes page.
NOTE: Episodes in CloudStream are not paginated, meaning that if you have a show with 21 seasons, all on different website pages you will need to parse them all.
A function can look something like this:
// The url argument is the same as what you put in the Search Response from search() and getMainPage()
override suspend fun load(url: String): LoadResponse {
val document = app.get(url).document
// A lot of metadata is nessecary for a pretty page
// Usually this is simply doing a lot of selects
val details = document.select("div.detail_page-watch")
val img = details.select("img.film-poster-img")
val posterUrl = img.attr("src")
// It's safe to throw errors here, they will be shown to the user and can help debugging.
val title = img.attr("title") ?: throw ErrorLoadingException("No Title")
var duration = document.selectFirst(".fs-item > .duration")?.text()?.trim()
var year: Int? = null
var tags: List<String>? = null
var cast: List<String>? = null
val youtubeTrailer = document.selectFirst("iframe#iframe-trailer")?.attr("data-src")
// The rating system goes is in the range 0 - 10 000
// which allows the greatest flexibility app wise, but you will often need to
// multiply your values
val rating = document.selectFirst(".fs-item > .imdb")?.text()?.trim()
?.removePrefix("IMDB:")?.toRatingInt()
// Sometimes is is not quite possible to selct specific information
// as it is presented as text with to specific selectors to differentiate it.
// In this case you can iterate over the rows til you find what you want.
// It is a bit dirty but it works.
document.select("div.elements > .row > div > .row-line").forEach { element ->
val type = element?.select(".type")?.text() ?: return@forEach
when {
type.contains("Released") -> {
year = Regex("\\d+").find(
element.ownText() ?: return@forEach
// Remember to always use OrNull functions
// otherwise stuff will throw exceptions on unexpected values.
// We would not want a page to fail because the rating was incorrectly formatted
)?.groupValues?.firstOrNull()?.toIntOrNull()
}
type.contains("Genre") -> {
tags = element.select("a").mapNotNull { it.text() }
}
type.contains("Cast") -> {
cast = element.select("a").mapNotNull { it.text() }
}
type.contains("Duration") -> {
duration = duration ?: element.ownText().trim()
}
}
}
val plot = details.select("div.description").text().replace("Overview:", "").trim()
// This is required to know which sort of LoadResponse to return.
// If the page is a movie it needs different metadata and will be displayed differently
val isMovie = url.contains("/movie/")
// This is just for fetching the episodes later
// https://sflix.to/movie/free-never-say-never-again-hd-18317 -> 18317
val idRegex = Regex(""".*-(\d+)""")
val dataId = details.attr("data-id")
val id = if (dataId.isNullOrEmpty())
idRegex.find(url)?.groupValues?.get(1)
?: throw ErrorLoadingException("Unable to get id from '$url'")
else dataId
val recommendations =
document.select("div.film_list-wrap > div.flw-item").mapNotNull { element ->
val titleHeader =
element.select("div.film-detail > .film-name > a") ?: return@mapNotNull null
val recUrl = fixUrlNull(titleHeader.attr("href")) ?: return@mapNotNull null
val recTitle = titleHeader.text() ?: return@mapNotNull null
val poster = element.select("div.film-poster > img").attr("data-src")
MovieSearchResponse(
recTitle,
recUrl,
this.name,
if (recUrl.contains("/movie/")) TvType.Movie else TvType.TvSeries,
poster,
year = null
)
}
if (isMovie) {
// Movies
val episodesUrl = "$mainUrl/ajax/movie/episodes/$id"
// Episodes are often retrieved using a separate request.
// This is usually the most tricky part, but not that hard.
// The episodes are seldom protected by any encryption or similar.
val episodes = app.get(episodesUrl).text
// Supported streams, they're identical
val sourceIds = Jsoup.parse(episodes).select("a").mapNotNull { element ->
var sourceId = element.attr("data-id")
if (sourceId.isNullOrEmpty())
sourceId = element.attr("data-linkid")
if (element.select("span").text().trim().isValidServer()) {
if (sourceId.isNullOrEmpty()) {
fixUrlNull(element.attr("href"))
} else {
"$url.$sourceId".replace("/movie/", "/watch-movie/")
}
} else {
null
}
}
val comingSoon = sourceIds.isEmpty()
return newMovieLoadResponse(title, url, TvType.Movie, sourceIds) {
this.year = year
this.posterUrl = posterUrl
this.plot = plot
addDuration(duration)
addActors(cast)
this.tags = tags
this.recommendations = recommendations
this.comingSoon = comingSoon
addTrailer(youtubeTrailer)
this.rating = rating
}
} else {
val seasonsDocument = app.get("$mainUrl/ajax/v2/tv/seasons/$id").document
val episodes = arrayListOf<Episode>()
var seasonItems = seasonsDocument.select("div.dropdown-menu.dropdown-menu-model > a")
if (seasonItems.isNullOrEmpty())
seasonItems = seasonsDocument.select("div.dropdown-menu > a.dropdown-item")
seasonItems.apmapIndexed { season, element ->
val seasonId = element.attr("data-id")
if (seasonId.isNullOrBlank()) return@apmapIndexed
var episode = 0
val seasonEpisodes = app.get("$mainUrl/ajax/v2/season/episodes/$seasonId").document
var seasonEpisodesItems =
seasonEpisodes.select("div.flw-item.film_single-item.episode-item.eps-item")
if (seasonEpisodesItems.isNullOrEmpty()) {
seasonEpisodesItems =
seasonEpisodes.select("ul > li > a")
}
seasonEpisodesItems.forEach {
val episodeImg = it?.select("img")
val episodeTitle = episodeImg?.attr("title") ?: it.ownText()
val episodePosterUrl = episodeImg?.attr("src")
val episodeData = it.attr("data-id") ?: return@forEach
episode++
val episodeNum =
(it.select("div.episode-number").text()
?: episodeTitle).let { str ->
Regex("""\d+""").find(str)?.groupValues?.firstOrNull()
?.toIntOrNull()
} ?: episode
// Episodes themselves can contain quite a bit of metadata, but only data to load links is required.
episodes.add(
newEpisode(Pair(url, episodeData)) {
this.posterUrl = fixUrlNull(episodePosterUrl)
this.name = episodeTitle?.removePrefix("Episode $episodeNum: ")
this.season = season + 1
this.episode = episodeNum
}
)
}
}
4. Loading links
This is usually the hardest part when it comes to scraping video sites, because it costs a lot to host videos. As bandwidth is expensive video hosts need to recuperate their expenses using ads, but when scraping we bypass all ads. This means that video hosts have a big monetary incentive to make it as hard as possible to get the video links.
This means that you cannot write just one piece of skeleton code to scrape all video hosts, they are all unique. You will need to customized scrapers for each video host. There are some common obfuscation techniques you should know about and how to detect them.
Obfuscation techniques to know about:
Base64:
This is one of the most common obfuscation techniques, and you need to be able to spot it inside documents. It is used to hide important text in plain view.
It looks something like this: VGhpcyBpcyBiYXNlNjQgZW5jb2RlIHRleHQuIA==
A dead giveaway that it is base64 or something similar is that the string ends with ==, something to watch out for, but not required. If you see any suspicious string using A-z in both uppercase and lowercase combined with some numbers then immediately check if it is base64.
AES encryption:
This is the more annoying variant of Base64 for our purposes, but less common. Some responses may be encrypted using AES and it is not too hard to spot.
Usually encrypted content is encoded in Base64 (which decodes to garbage), which makes it easier to spot. Usually sites are not too covert in the use of AES, and you should be alerted if any site contains references to enc, iv or CryptoJS. The name of the game here is to find the decryption key, which is easiest done with a debugger. If you can find where the decryption takes place in the code, usually with some library like CryptoJS then you can place a breakpoint there to find the key.
More to come later!