Everything you own to read or watch, in one place — and it never forgets where you stopped. Manga, comics, films, series and anime, served from your own disk, with one "continue" shelf for all of them and an API for your tools.
Plan · Security · En castellano · MIT
Status: 0.7. It reads, plays, remembers and takes uploads, with a profile for each person, posters and details from TMDB or AniList if you turn them on, a year in review, OPDS for phones and an API and MCP server for tools. Spanish UI first. The plan is complete; what changed is in the changelog.
A self-hosted library for what you already have on disk:
- A reader for manga and comics (CBZ, ZIP, PDF, folders of images): right-to-left, left-to-right, vertical and double-page. It remembers the volume, the chapter and the page.
- A player for films, series and anime: what the browser plays directly, the rest remuxed on the fly with ffmpeg, subtitles included (beside the file or inside it). Pick the audio track and the subtitles; each profile's choice carries to the next episode. It remembers the episode and the second.
- Posters and details: a
poster.jpgin the work's folder, a frame you pick, or, if you turn it on, the poster, synopsis, genres and credits from TMDB (films and series) or AniList (anime and manga). - One library: the same story as a manga and as an anime is one work, with one place to continue.
- Profiles for everyone in the house, picked at the door like on a streaming service: each with its own avatar, its own "continue", its own phone catalog and its own year in review (time watched, pages read, what they finished).
- An API with a token, so a tool can list works, ask for a page or a frame, and read where you are. An MCP server for agents, too.
It never fetches works: no downloader, no scraper, no source list. It reads the files you put in its folders. See SOUL.md for the rules it keeps.
library/
video/<work>.mp4 + <work>.yaml a film (its poster: <work>-poster.jpg)
video/<work>/S01E01.mkv … + work.yaml a series or an anime (poster.jpg in the folder)
pages/<work>/vol-01.cbz … + work.yaml a manga or a comic (poster.jpg in the folder)
The yaml can also say what the work is about, and that always wins over a lookup:
overview, genres, and director, author or studio.
The folders are the truth: copy them with rsync, back them up without archivist running. The database only holds what a disk cannot — where you are.
Docker
docker build -t archivist https://github.com/legiosai/archivist.git
docker run -d --name archivist -p 8780:8780 \
-v /path/to/library:/library -v archivist-data:/data \
-e ARCHIVIST_TOKEN="$(openssl rand -hex 24)" archivistFrom source (Node 22 or newer, plus ffmpeg and poppler-utils)
git clone https://github.com/legiosai/archivist && cd archivist
npm ci && npm run build
ARCHIVIST_LIBRARY=/path/to/library ARCHIVIST_TOKEN=… npm startA systemd user unit and an example environment file are in deploy/.
| Variable | Default | |
|---|---|---|
ARCHIVIST_LIBRARY |
— | The folder with video/, pages/ and stills/. |
ARCHIVIST_DATA |
./data |
The database (progress) and the caches. |
ARCHIVIST_HOST |
127.0.0.1 |
Any other address requires a token. |
ARCHIVIST_PORT |
8780 |
|
ARCHIVIST_TOKEN |
— | For the API (Authorization: Bearer) and the browser login. |
ARCHIVIST_BACKUP_DIR |
— | A daily copy of the database (progress), the newest 14 kept. |
ARCHIVIST_OWNER_NAME |
Yo |
The first profile's name. More profiles are added in the UI. |
ARCHIVIST_TRUSTED |
— | Networks that need no token: loopback, tailscale, lan, or CIDRs (192.168.1.0/24). Judged by the connection's address, never by headers, and never for a request that came through a reverse proxy. |
ARCHIVIST_HOSTS |
— | More host names that reach it from a trusted network (beyond addresses, localhost, one-word names and .ts.net/.local/.lan). Anything else needs the token, which stops DNS rebinding. |
ARCHIVIST_PROXIES |
loopback |
Reverse proxies whose CF-Connecting-IP / X-Forwarded-For is believed for the client's address. In Docker, the proxy's network (172.17.0.0/16). Any request with those headers is untrusted either way. |
ARCHIVIST_PROXIED_VIDEO |
off | Serve video to requests that came through the public proxy. Off by default: at home, video goes over the LAN or Tailscale. |
ARCHIVIST_METADATA |
— | Where to look works up: tmdb, anilist, or both. Off by default: then nothing leaves the server. When on, the server (never the browser) sends a work's title and year once, and keeps what comes back. |
ARCHIVIST_TMDB_TOKEN |
— | TMDB's API read access token (or a v3 API key), free at themoviedb.org. AniList needs none. |
Everything the UI does goes through the API: works, units, pages, single frames,
progress, events and resumable uploads. Tools get pages and frames, never clips. It is
described in OpenAPI 3.1 at /api/v1/openapi.json.
archivist is an MCP server at /mcp (Streamable HTTP, the same token). An agent can search the
library, open a work, look at a page or a frame, and read or save where someone is:
claude mcp add --transport http archivist https://your-server/mcp --header "Authorization: Bearer $ARCHIVIST_TOKEN"Tools: search_library, list_works, get_work, get_progress, set_progress, get_page,
get_frame, recent_events, list_profiles. /llms.txt sums it up for a model.
Put it behind a reverse proxy that terminates TLS (a Cloudflare Tunnel, Caddy, nginx). archivist
notices a proxied request (it carries CF-Connecting-IP, X-Forwarded-For or Forwarded; the
client's address is read from a proxy listed in ARCHIVIST_PROXIES, loopback by default) and
then: never treats it as a trusted network, logs
each wrong token with the real client address (archivist: auth failure from <ip>, for
fail2ban or CrowdSec), answers 429 after 10 wrong tokens in 15 minutes, marks the session cookie
Secure, sends HSTS, and keeps video at home unless ARCHIVIST_PROXIED_VIDEO=on. Reading,
progress and the API work as at home. See SECURITY.md.
The page has a web manifest and icons: "Add to home screen" puts archivist there. To open it as
its own app, without the browser's bar, browsers ask for HTTPS; on Tailscale, tailscale serve
gives the machine an https://<name>.<tailnet>.ts.net address.
http://your-server:8780/opds is an OPDS catalog: add it to Panels, Chunky, KOReader or Mihon.
Volumes stream page by page, and reading there moves the same "continue" shelf. For someone
else's profile, …/opds?profile=sol.
When ARCHIVIST_METADATA is on, posters, synopses, genres and credits come from
TMDB and AniList. This product uses the TMDB
API but is not endorsed or certified by TMDB.
npm install
npm test
npm run typecheck
npm run dev # the UI with hot reload, against a server on :8780