minihoard

A Rust CLI and MCP server to fetch your MyMiniFactory library, unpack and reorganize it into tidy per-creator/month folders, and repack it (tar.zst or zip, splittable) for archival and cataloging alongside stl-pack.

Browse and download from the terminal, or talk to an assistant (via the MCP server) — "show me June's Tribe releases", "download these".

Install

macOS

curl -fsSL https://github.com/irongollem/minihoard/releases/latest/download/install.sh | sh

Windows (PowerShell)

irm https://github.com/irongollem/minihoard/releases/latest/download/install.ps1 | iex

Both scripts download minihoard and minihoard-mcp, place them in a local bin folder, and add it to your PATH automatically.

Or build from source — needs Rust, plus CMake and (on Windows) NASM for BoringSSL: cargo build --release.

MCP (Claude Desktop)

After install, one command wires up the MCP server:

minihoard setup-mcp

That's it — it edits claude_desktop_config.json automatically. Restart Claude Desktop, then ask it to browse and download your library.

The MCP exposes status, config, list_library, preview_download, download_objects, pack, unpack, tidy, and job_status — so you can browse, download, repack, restore, and tidy entirely by chat. The long-running tools (download_objects, pack, unpack, tidy) run in the background: they return a job id right away and the assistant polls job_status to report live progress ("3 of 5 done, downloading Behir…") instead of blocking silently — which would otherwise trip the client's tool-call timeout on a big batch.

Setup

minihoard configure    # API client id (+ secret) and your library folder
minihoard login        # browser OAuth — opens a tab, stores a refresh token
minihoard sync-cookie  # import your MMF session cookie from your browser

The session cookie is needed only for library listing (downloads use OAuth). The easy way is minihoard sync-cookie, which reads it straight from a browser you're already logged into MyMiniFactory with (like yt-dlp --cookies-from-browser) — no manual copying. It reads the browser's on-disk cookie store; it does not drive your browser and needs no extension. It auto-detects across installed browsers, or pick one with --browser firefox. On macOS it prompts once for Keychain access to decrypt Chromium cookies (Firefox isn't encrypted, so no prompt). Re-run when the cookie expires (rarely — REMEMBERME is long-lived).

Windows + Chrome/Edge/Brave: recent Chromium versions (Chrome/Edge 127+) added app-bound encryption that ties cookie decryption to the browser itself, which breaks disk-based readers like this one. So on a recent Windows machine sync-cookie --browser edge (or chrome/brave) will likely fail. Two reliable options there:

• minihoard sync-cookie --browser firefox — Firefox doesn't encrypt cookies, so it always works. (Log in to MyMiniFactory in Firefox once.)
• minihoard set-cookie — paste the cookie: header from your browser's DevTools (Network tab → any www.myminifactory.com request). Works in any browser.

Secrets are stored in a 0600 file in the app config dir, never a system keychain, so headless/automated use never prompts.

Commands

minihoard list --month 2026-06            # filter by release month
minihoard list --creator "one page rules" # by creator
minihoard list --search dragon --undownloaded
minihoard list --source tribe             # by source channel (tribe/purchase/kickstarter/…)
minihoard get 806054 806051               # by id: download, unpack, clean, reorganize
minihoard get "dragon knight"             # by name: search, pick, then the same
minihoard get --month 2026-06             # batch: this month's whole drop (asks first)
minihoard get --month 2026-06 -j 6        # ...downloading 6 in parallel
minihoard get --creator "one page rules" --undownloaded   # batch: only what's new
minihoard get 806054 --keep-archive       # ...but keep the original .zip
minihoard get --month 2026-06 --pack --split 4G --name "Dungeon Classics - 2026-06"
                                          # download, then repack the month into named 4 GB chunks
minihoard pack ~/mmf/Creator-06-2026      # repack a folder for backup (tar.zst)
minihoard pack DIR --format zip           # ...as a broadly-supported .zip
minihoard pack DIR --split 4G             # ...split into 4 GB volumes (tar.zst)
minihoard pack DIR --name "Archive Name"  # ...with a custom archive filename
minihoard tidy                            # tidy whole library (strip junk, collapse nesting)
minihoard tidy ~/mmf/some-release         # ...or specific folders
minihoard unpack FILE.zip                 # restore a .zip or .tar.zst archive
minihoard unpack FILE.tar.zst.001         # ...or a split archive (first volume)
minihoard unpack FILE.tar.zst --delete-archive  # ...and remove it once extracted
minihoard where                           # show config + where files are stored
minihoard whoami                          # show the logged-in account
minihoard upgrade                         # update to the latest release

get (alias of download) is the one-shot command: give it object ids or a name to search for, and it produces ready-to-use releases — each object is unpacked into <unpack_dir>/{creator}-{MM-YYYY}/{release}/, macOS artifacts (__MACOSX/, .DS_Store, ._*) are stripped, and the .zip is deleted (unless --keep-archive). A name that matches several items lets you pick which to fetch. (Searching by name needs the session cookie; ids don't.) list marks items already downloaded; downloads are tracked in a local manifest.

Redundant nesting is collapsed automatically: if an archive contains nothing but a single top-level folder (the common Release/Release/files case), that wrapper is removed and the release folder takes the inner folder's name — so you get {creator}-{MM-YYYY}/{release}/files, not a doubled-up path. (unpack does the same.)

For the monthly drop, batch a whole set with the same filters list uses — --month, --creator, --search, --source, --undownloaded. get previews the matches and asks before downloading (use -y to skip the prompt). For example, minihoard get --month 2026-06 --undownloaded grabs everything new this month, and minihoard get --source tribe --month 2026-06 grabs just your Tribe-tier releases. list with no filters shows breakdowns by month and by source to help you choose.

Packing for backup

pack turns a clean release folder into a single archive for off-site backup. Two formats:

• tar.zst (default) — best compression and speed, and the only format that can be split into fixed-size volumes (--split 4G) for chunked backup. It's a stream, so reading one file means decompressing the archive. Restore with minihoard unpack, or with standard tools (zstd -dc a.tar.zst | tar -x; for splits, cat a.tar.zst.* | zstd -dc | tar -x).
• zip (--format zip) — broadly supported, native double-click extraction, and random-access (a catalog can read one entry without unpacking everything). Single file only — no --split.

--level N sets the zstd level (1–22, default 19). Compression runs multi-threaded across your cores. --name "…" sets the archive's base filename verbatim (for a strict archive naming convention) instead of the folder name; the archive's internal layout still keeps the real folder name.

Each archive gets a <archive>.json sidecar index next to it (disable with --no-sidecar). It lists the creator, release month, and every file inside — categorized as model / image / doc / other — so a catalog (or you) can see what's in an archive, and find printable meshes or preview images, without decompressing it. This matters most for tar.zst, which has no random access.

To restore, minihoard unpack handles .zip and .tar.zst (point it at the .001 volume for a split set). Add --delete-archive to remove the archive — all split volumes plus the sidecar — once extraction succeeds.

How it works

• Auth: OAuth2 Authorization Code via a localhost loopback redirect.
• Discovery: the website's data-library/objectPreviews endpoint lists the whole library (id, name, creator, release month, added date).
• Download: object files via the OAuth API; the download host is behind Cloudflare, so requests use a Chrome-impersonating client to pass the bot check, with resumable streaming. Objects download with bounded concurrency (default 5, like a browser's per-host connections; tune with -j or the download_concurrency config key).

get --pack chains the repack onto a download: after the releases land, each touched month group is packed into a tar.zst archive. --split/--name (which both imply --pack) set the chunk size and the archive's filename, so a whole monthly drop becomes one named, chunked backup in a single command.

License

MIT — see LICENSE.