Privacy and security¶
What MangaPixer keeps to itself, what it can send when an admin allows it, and how to report a vulnerability.
- No telemetry, analytics or phone-home. Fonts and icons are bundled and served by your own server, never from a CDN.
- Three optional internet features, all off by default and switched on only by an admin: the Update Checker, fetching series information from the allowed sites - MangaUpdates, the Grand Comics Database (comics), MangaDex (volume covers, and series MangaUpdates does not list), AniList (chapters per volume) and Wikipedia (volume lists) (see What leaves your server) - and, from 1.43.0, tracker sync to the users' own MyAnimeList and MangaUpdates lists (see Tracker sync).
- Ratings and recommendations stay on your server. Ratings are shared only inside one server, only once an admin turns on Shared ratings (1.53.0), and only as an average and a count - never who rated what; a rating in a user's Private library never counts, and each user can stop sharing in Settings. Recommended for you and More like this are worked out on your server from your own ratings, favorites and reading; nothing is sent anywhere.
- Logs never contain paths or titles, only IDs, counts, timings and sanitized error codes. Reader-facing API responses never contain filesystem paths.
- Source media is read-only, enforced by the code and by the
:romounts. - No default credentials. The first admin is created by you on the setup screen, and that endpoint refuses once any user exists.
- API tokens (1.33.0) are created only by an admin, read only the metadata export - and, only when the admin ticks it (1.36.0), may request a full library scan - and nothing else, and are stored as a SHA-256 fingerprint - the token itself is shown once and never written to the database or the logs (see API tokens).
- Session cookies use ASP.NET Core Data Protection; the keys live in
<DataRoot>/keys, so treat the data volume as sensitive.
What leaves your server¶
Out of the box, nothing: MangaPixer makes no internet requests until an admin turns something on. The Update Checker asks GitHub once a day whether a newer release exists. Fetch series information from the web (account menu > Metadata Manager, after a consent text, and per library) lets an admin look a folder up on MangaUpdates: only the search text the admin confirms in the Identify dialog (or, from 1.38.0, ticks under Search the web for the rest in Match folders by name), MangaUpdates record numbers, a fixed list of types to leave out (when Hide doujinshi & novels is ticked), from 1.38.0 the MangaUpdates author numbers of the creators of stored series (see MangaUpdates author pages), from 1.45.0 the name the admin confirms when looking for an artist (see MangaUpdates author search) and the author numbers of the artists an admin picked there, and the fixed User-Agent (see User-Agent) are sent, only to api.mangaupdates.com and cdn.mangaupdates.com, and only when an admin presses Search, Look up, Preview, Link, Accept, Refresh, Look up the rest or Search (Match folders by name). File paths, your file list, user accounts and reading progress are never sent. Fetched data and cover art are stored on your server and served from it, so readers' browsers never contact MangaUpdates. (New in 1.44.0: in Metadata Manager's review lists, an admin can click a series' title to open its page on MangaUpdates or MangaDex - the admin's browser opens that page in a new tab, without telling the site which address it came from; MangaPixer itself makes no request for it.) Automatic matching is a separate switch with its own consent text, off by default: when an admin turns it on, libraries that fetch from the web are matched in the background, so the cleaned name of each new series folder (or of an archive that is its own work: in a collection or artist's folder, or loose next to other folders; from 1.34.0 also each work inside a folder an admin marked Collection about a linked series, then searched as <series> dj - <title> with the series' MangaUpdates title - never a folder or file name - in front) is sent without anyone reviewing it first (and, when the first page of results is a tie or has nothing close enough, the second page of that same search is read), and linked series are refreshed by their MangaUpdates number. When two records tie on the title for a folder of volumes or a one-shot, it may also download those two records' cover images from cdn.mangaupdates.com to compare them with the folder's own cover thumbnail on your server (Compare covers, part of Automatic matching; the downloaded covers are deleted right after, and nothing from your library is sent with them). It uses the same hosts and the same daily request budget, at most one request per second, and never looks inside a folder marked Don't match. The cleaned name of a folder that is still waiting can be sent again: an unmatched one after 30, 90 and 180 days, and one in Needs review once after an update changes how matches are scored (never one an admin linked, confirmed or marked Don't match); this is the same request to the same site. Metadata__NetworkDisabled=true switches all of this off regardless of the admin settings. Details: Series information.
Series data fetched from the web is provided by MangaUpdates and credited to it wherever it is shown.
Allowed sites and AniList¶
New in 1.28.0. MangaPixer contacts only the allowed sites listed in Metadata Manager > Settings: MangaUpdates, the Grand Comics Database (1.32.0, see below), MangaDex (1.29.0, see below), for chapters per volume AniList, and Wikipedia (1.32.0, see below). An admin can remove any of them; a removed site gets no request of any kind. AniList is asked when an admin presses Get chapters per volume from AniList or Chapters per volume (AniList) in the Missing report, and - from 1.29.0, with Automatic matching on - in the background for a linked series that MangaDex gives no volume list for; only for libraries whose Fetch switch is on, and never to match a folder: MangaPixer sends the AniList record number when it is known (entered by an admin, stored, or taken from the linked MangaDex record's own AniList link), otherwise the MangaUpdates title of a series that is already linked (never a folder or file name), a fixed "manga, not novels" filter and the fixed User-Agent, only to graphql.anilist.co. It stores the matching entry's volume and chapter totals on your server, counts every request in the same daily budget, sends at most one request per second, and waits when AniList asks it to slow down. When an update changes the allowed sites or what is sent, fetching stays off until an admin accepts the new consent text (After an update: accept again). AniList data is credited to AniList where it is shown.
MangaDex: volume covers, volume lists and series records¶
New in 1.29.0. MangaDex is used in two ways. (1) As a companion of a series already linked to a MangaUpdates record, for that series' volume -> chapter list and its volume covers: MangaPixer sends only the title or an associated title of the linked MangaUpdates record (never a folder or file name). (2) New in 1.42.0: as the series record of a work MangaUpdates does not list - when an admin picks MangaDex in Identify (the search text the admin confirms, or a pasted MangaDex title address - then only its id), and with Automatic matching as described below; a series linked to MangaDex is refreshed by its MangaDex id and is its own companion for volume lists and covers. Requests send only that text or title, MangaDex record numbers, a fixed content-rating list, fixed includes (cover art, authors, artists), your preferred and the series' original cover language as a filter (your preferred language also for the volume -> chapter list, to learn which chapters are released in it - from 1.43.0 read a second time without the chapters MangaDex took down, to learn which of them can actually be read there; the same record number and language, nothing more), and the fixed User-Agent (no Via header, no cookies, nothing that identifies your server), only to api.mangadex.org and uploads.mangadex.org. A MangaDex record is used only when its own MangaUpdates link names the linked record, or when an admin chose it (Change MangaDex match…). Cover images are downloaded by the record number and file name MangaDex returned (the 512-pixel version), re-encoded and hashed by MangaPixer's media worker and stored in the data root; readers' browsers never contact MangaDex. From 1.33.0 MangaPixer also keeps the official-source links listed in the same MangaDex record answer (the official original-language and English pages, and the BookWalker, Amazon, ebookjapan and CDJapan store pages) for the metadata export; it never requests those addresses itself, and reading them adds no request.
With Automatic matching on, a work that neither MangaUpdates nor (for comics) the Grand Comics Database found anything close for may be searched on MangaDex by the same cleaned name (at most two searches; never for comics or inside a Collection about folder; only while Search MangaDex when MangaUpdates finds nothing is on) - only to propose a match under Needs review. Doujinshi are left out on your server, not by a filter sent to MangaDex. A candidate's cover (the 256-pixel version) is shown from MangaPixer while an admin chooses, kept in memory for an hour; a linked MangaDex series' main cover is stored as its poster.
With Automatic matching on and Volume covers from the web on (Metadata Manager > Settings > Volume covers; on by default; Metadata__AutoMatch__VolumeCovers=false switches it off regardless of the setting), this happens in the background for every linked series: find its MangaDex record, read its volume list (in all languages and in your preferred language) and its volume covers in the two languages, download the covers of volume 1 and of the volumes the folder holds (as volume files, as all of their chapters by MangaDex's volume list, or as an estimated ~ Volume N stack of the Volumes view) - or, when MangaDex lists no volume 1 cover (webtoons, for example), its main cover - and look again on the series' refresh schedule until a cover in your preferred language appears. Without Automatic matching, MangaDex is contacted only when an admin asks (Refresh, Change MangaDex match…). Everything counts in the same daily budget, at most one request per second, and pauses while MangaDex asks to slow down. Nothing inside a folder marked Don't match is looked up. Removing MangaDex from the allowed sites stops every request to it; covers already stored stay and keep being shown until you use Delete stored volume covers. Logs record IDs, counts, status codes and timings only. Cover images and volume data come from MangaDex, credited under its setting in Metadata Manager.
Grand Comics Database: comics and graphic novels¶
New in 1.32.0. The Grand Comics Database (GCD) is the site MangaPixer asks about comics and graphic novels - works with a comics signal (for example a folder declared Comic or Graphic novel, a folder below a Comics / BD folder, comic-issue or album numbering together with another comics sign, or a comics link inside ComicInfo) - and any work an admin chooses to search there in Identify. It is sent only the search text the admin confirms (with Automatic matching on, also the cleaned name of a new comics folder - first to GCD, and to MangaUpdates only when GCD finds nothing close), a start year when the name has one, GCD record numbers and the fixed User-Agent; no account or key, nothing that identifies your server, only to www.comics.org and files1.comics.org. GCD's cover thumbnails are shown only while you choose, and may be compared in memory with a folder's own cover; they are never stored as covers - comic files keep their own cover. GCD answers anonymous users slowly, so MangaPixer sends it at most 25 requests an hour, inside the same daily budget. Comics data is credited "Grand Comics Database (comics.org), CC BY-SA 4.0" where it is shown.
Wikipedia: volume lists¶
New in 1.32.0. Wikipedia is used only as a companion of a series already linked to a MangaUpdates record, to read which chapters make up each volume - with each volume's English release date and ISBN - from its English "List of ... chapters" page, where MangaDex's list has gaps; it is never used to identify or match a folder. MangaPixer sends only the linked MangaUpdates record number (to Wikidata, to find the English article linked to it), the page titles found that way or the linked series' MangaUpdates title (never a folder or file name), fixed query parameters and the fixed User-Agent, only to en.wikipedia.org and www.wikidata.org, one request at a time. Only volume and chapter numbers, dates, ISBNs, the page title and its revision are kept, credited to Wikipedia with a link. Without Automatic matching it is contacted only when an admin asks; in the background it runs only while Volume covers from the web is on.
MangaUpdates author pages: artists' other names¶
New in 1.38.0. When an admin presses Look up the rest under Artists' other names (Metadata Manager > Settings), MangaPixer reads the MangaUpdates author page of each creator that a stored, linked MangaUpdates record names with an author number - in libraries whose Fetch switch is on - to learn that artist's other names (pen names, spellings, scripts) for Match folders by name. Each request is GET https://api.mangaupdates.com/v1/authors/<number> with only the author number and the fixed User-Agent: never a name, a search, a path or anything else from your library. It runs only when an admin asks - never in Automatic matching, the background refresh or any other background work - one request per second, counted in the daily budget, and stops when the budget is spent, when MangaUpdates asks MangaPixer to slow down, or when web lookups are switched off. An author is asked again only after 180 days. Only the names are stored on your server; Delete all fetched web data removes them. Because this is a new kind of request, the consent text of Fetch from the web changed in 1.38.0 (version 5): fetching stays off after the update until an admin accepts it again. Logs record IDs, counts, status codes and timings only.
MangaUpdates author search: finding an artist¶
New in 1.45.0. An admin can look an artist up on MangaUpdates by name, in exactly two places: Match folders by name as artist folders (under Search MangaUpdates for the rest, off until ticked: each ticked folder's name as listed - editable - for the folders without a match and the artist folders without a MangaUpdates author yet) and Find on MangaUpdates in the Artist folder dialog (the Artist field's text, once per press). Each request is POST https://api.mangaupdates.com/v1/authors/search with exactly {"search": "<the text>", "page": 1, "perpage": 10} and the fixed User-Agent: never a path, a file list, an archive name, a declared fact or anything about users. It is never automatic and never runs in the background; Match folders by name sends at most 50 per run, one per second, and stops when the budget is spent, when MangaUpdates asks MangaPixer to slow down, or when web lookups are switched off. Every request counts in the daily budget; repeating the same text within an hour sends nothing. The authors found (name, the other name the search matched, number of series, up to three genres, a link to the author's MangaUpdates page - opened by the admin's own browser without telling MangaUpdates where it came from; no images) are only offered and are not stored. Picking one and pressing Mark sends nothing: the folder's declared artist keeps the author's MangaUpdates number, an admin decision that Delete all fetched web data keeps. That number joins the authors whose other names an admin can look up under Artists' other names. Because this is a new kind of request, the consent text of Fetch from the web changed in 1.45.0 (version 7): fetching stays off after the update until an admin accepts it again. Logs record IDs, counts, status codes and timings only - never the search text or a name.
MangaDex and MangaUpdates: noticing new chapters¶
New in 1.45.0. With Automatic matching on (and its consent, and Fetch from the web with consent version 7), MangaPixer checks for new chapters every few hours: it sends MangaDex the MangaDex record ids of your linked series' MangaDex records - 100 per request, with the fixed content-rating list and page size, every 6 hours by default (an admin may choose every 1 to 6 hours in Scheduled jobs) - to read each record's latest uploaded chapter; and it reads MangaUpdates' public list of the day's releases up to four times a day, sending only the page number. Nothing from your library is sent - no names, paths or files. The releases are compared on your server with the series you have linked; of them MangaPixer keeps only which of your series had a release, never a title. Per MangaDex record it keeps the latest chapter's id and when it changed (shared in the metadata export). These requests count in the daily request budget and are paced like every other.
API tokens¶
New in 1.33.0. An admin can create API tokens in Administration so another program - MangaList - can read the metadata export (/api/v1/export/). A token can only read that export: it is refused for anything but reading requests, and sent to any other page it is ignored as if it were absent. From 1.36.0 an admin can also tick Request library scans for a token: it may then ask for a full scan of a library (POST /api/v1/export/libraries/{id}/scan, at most once per library every 5 minutes) - the same local scan as Scan now, which changes no files, links or settings and sends nothing to the internet. That one address accepts only tokens, never a browser sign-in. Each started scan is recorded in the Audit trail. It works only while the admin who created it is an active admin, and until it is revoked or expires. MangaPixer stores only a SHA-256 fingerprint of each token; the token is shown once, never logged (log lines name a token by its public id), and never sent anywhere. Use HTTPS when the program reaches MangaPixer from outside your home network. Details: API tokens.
Tracker sync¶
New in 1.43.0. Off until an admin allows a tracker in Administration > Trackers (after its own text - MyAnimeList only on a personal, self-hosted server or a personal desktop install, with an app this server's admin registered) and a user connects their own account on their Tracking page (account menu > Tracking, after the user text). Then, while that user reads, the server reads that user's entry for a linked series and, only when MangaPixer is ahead, updates it: the tracker's own number of the series (the linked MangaUpdates record's number; for MyAnimeList the number the series' MangaDex record names, or one an admin entered - never found by a title search), chapters and volumes read, a status (Reading, or Completed / Complete only for a finished series at the tracker's total), for MyAnimeList the start and finish dates (the day the user first read the series here and the day it was completed, or dates the user set on the Tracking page; MangaUpdates gets no dates), the user's sign-in token, and the fixed User-Agent, to api.myanimelist.net (and myanimelist.net when signing in) or api.mangaupdates.com. Never titles, file or folder names, paths, which files were read, page positions, other users or anything that identifies your server. New in 1.53.0: your own score is sent only if you turn on Keep my score in step with my MangaPixer rating for that tracker - your rating of a linked series folder when you set, change or clear it, only for a series already on your list there (never a single file's rating; a score never adds a series); Take my scores only reads your scores there. This needed new consent texts (version 2): syncing pauses until the admin and each user accept them. Folders marked read at once and imported reading wait for the user's Send; Private libraries and doujinshi are not synced unless the user turns sync on for that folder or series; a number on the tracker is never lowered without the user's click. Nothing is sent while nobody reads, except what a user starts (Sync now, the two Set progress buttons, for one series or for series you selected - from the tracker to MangaPixer only reads the tracker and marks read on your server, never unmarks -, Set tracker's volumes to N - which writes only the tracker's volume count, raised to what its own chapter count completes on the series' volume list from MangaDex or Wikipedia, with the entry's own status (new in 1.44.0) -, Check my list - which only reads). The MangaUpdates password is sent once to get a token and never stored; the MyAnimeList password is typed only on myanimelist.net. Tokens are stored encrypted with the server's keys, never shown, logged or exported, and deleted on disconnect; each user makes at most a daily number of requests per tracker that the admin sets (2000 by default). The browser itself only opens myanimelist.net to sign in, and a series' page on the tracker when you click its link (a new tab, no referrer). Trackers__NetworkDisabled=true switches it off. Details: Tracker sync, Tracker setup.
User-Agent¶
New in 1.32.0. Every request to an allowed site carries the same User-Agent: MangaPixer/<version> (+https://github.com/dixit92/mangapixer) - the product, its version and the project page, as MangaDex and Wikimedia ask. It is identical on every MangaPixer server of that version, so it does not identify yours. (Before 1.32.0 it was MangaPixer-Metadata.) The Update Checker keeps its own generic User-Agent for GitHub.
To report a vulnerability, see SECURITY.md.