Memory Toast Make Card
smallseven1213/memory-toast-make-card
Cria decks de flashcards Memory Toast com imagens geradas por IA
Ver no GitHubSkill para Claude Code que transforma materiais de estudo (PDFs, notas, pesquisas) em decks de flashcards para Memory Toast, com geração opcional de imagens via OpenAI ou Gemini, e upload automático para sua conta.
Instalação
npx convez-ai smallseven1213/memory-toast-make-card1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Criar flashcards para idiomas com imagens customizadas
- Converter notas de estudo em decks estruturados
- Gerar materiais de revisão com ilustrações automáticas
Veredito

Ferramenta bem-engineered que automatiza a criação de flashcards com suporte a geração de imagens. Requer conta Memory Toast e chaves de API próprias (OpenAI/Gemini), ideal para quem precisa escalar produção de material didático.
README
A tradução para o português deste README ainda está na fila.
memory-toast-make-card
A Claude Code skill that turns your study material — requirements, web research, PDFs, images, notes — into Memory Toast flashcard decks (卡包), optionally generates card images with your own OpenAI/Gemini key, and uploads the deck to your Memory Toast account as a ZIP pack.
It is the public, token-authenticated counterpart of the internal make-card skill: you sign
in once with your own account, and your password is never stored — only a rotating refresh token.
Install
Option A — npm
# Into your user skills (~/.claude/skills/)
npx memory-toast-make-card install
# Or into the current project (./.claude/skills/)
npx memory-toast-make-card install --project
Option B — skills.sh
Install from the registry, or copy the folder into ~/.claude/skills/memory-toast-make-card/
manually. The skill is the directory containing SKILL.md.
Prerequisites
- A Memory Toast account — the same one you use in the app.
- Log in once (stores a refresh token in
~/.memory-toast/credentials.json, chmod 600; no secret is written to disk except the rotating token):# Email/password account: python3 ~/.claude/skills/memory-toast-make-card/scripts/mt_login.py # Google/Facebook account (no password): copy the token from the app # (Settings → Copy upload token), then paste it here: python3 ~/.claude/skills/memory-toast-make-card/scripts/mt_login.py tokenmt_login.py whoamishows the logged-in account;mt_login.py logoutremoves the token. - (Optional) image generation — export your own key for the provider you want:
You always supply your own key; the skill never provides one. Image generation calls a paid API — the skill confirms the cost with you before generating in bulk.export OPENAI_API_KEY=sk-... # OpenAI gpt-image-1 export GEMINI_API_KEY=... # Google Imagen / Gemini
Usage
In Claude Code, just ask — for example:
Make a Memory Toast deck of 30 common Japanese N3 verbs, each with a flat-vector icon image.
The skill will gather requirements, draft a few sample cards for your approval, generate any images (with your confirmation on cost), build the ZIP, and upload it. Then open the deck in the Memory Toast app and tap Download on the "new version available" banner.
You can also drive the scripts directly:
S=~/.claude/skills/memory-toast-make-card/scripts
# One-time: choose where decks are kept (remembered; keeps them out of /tmp)
python3 $S/mt_config.py set deck-root ~/Documents/MemoryToast/decks
# Generate an image with your own key
python3 $S/gen_image.py --provider openai \
--prompt "flat vector icon of a person eating, warm palette, white background" \
--out my-deck/assets/taberu.png
# Validate a deck directory offline (builds my-deck/build/pack.zip)
python3 $S/upload_pack.py my-deck --dry-run
# Create a new deck and upload (writes my-deck/.memory-toast.json)
python3 $S/upload_pack.py my-deck
# Update the SAME deck later — just run it again; it auto-reads the record
python3 $S/upload_pack.py my-deck
# Publish to the public Library, then release new versions later
python3 $S/library_pack.py publish my-deck --category language --description "..."
python3 $S/library_pack.py release my-deck --changelog "Added 10 cards"
See references/pack-format.md for the deck.json schema, the
upload protocol, limits, and version/conflict rules.
Deck storage, updates & publishing
- Decks are built under a remembered root (
mt_config.py set deck-root <path>), not /tmp, so the editable source survives reboots. Each deck folder gets a.memory-toast.jsonrecord (deckId, version, structure, …) on upload — the state a later session reads to continue. - To update a deck, edit it and re-run
upload_pack.py <deck-dir>; it reads that record and bumps the server version automatically — no ids to remember (--newforces a fresh deck). - To share a deck,
library_pack.py publishit to the public Library (categories: language, science, history, programming, math, geography, exam, other), thenreleasenew versions as you update it.statuslists what you've published.
Rich text in cards
A card's frontContent / backContent (and a section's caption) may contain a small HTML
subset for formatting; plain text is always valid, so existing decks keep working unchanged.
make-card emits this HTML straight from deck.json — what you write is what the app renders.
Allowed tags:
<b>/<strong>— bold<i>/<em>— italic<u>— underline<br>— line break<span style="font-size:sm|base|lg|xl;color:<token>">— inline size and/or color<p style="text-align:left|center|right">…</p>— paragraph alignment
Font sizes are four semantic levels — sm, base, lg, xl (not arbitrary px). Color tokens
are: primary, red, orange, green, blue, purple, gray.
{ "frontContent": "<b>紅蘋果</b>", "backContent": "apple <span style=\"font-size:lg;color:red\">(fruit)</span>" }
Anything outside the whitelist (other tags, attributes, styles, or color values) is dropped
while its text is kept — no scripts ever run. See
references/pack-format.md §3.1 for the full rules.
Configuration
- Server URL resolves as:
--apiflag →MEMORY_TOAST_API_URLenv → stored value → built-in default. Self-hosters can point the skill at their own server without editing files. - Image model is overridable with
--model. Defaults:gpt-image-1(OpenAI),imagen-4.0-generate-001(Gemini). Pass agemini-*-imagemodel (e.g.gemini-2.5-flash-image) to use thegenerateContentpath instead of Imagenpredict.Note: Google's older image-generation model endpoints are deprecated after 2026-06-30 — switch the default via
--modelif generation starts failing.
Security notes
- The password is read interactively (never echoed) and never stored — only the rotating
7-day refresh token, in a
chmod 600file under your home directory.logoutdeletes it. - API keys and tokens are never printed; error paths redact them.
- These are stateless JWTs: a leaked refresh token cannot be individually revoked before
its 7-day expiry. Keep the credential file private and run
logouton shared machines. - An upload replaces the deck's server pack wholesale — push any un-synced phone edits first (see pack-format.md §5).
Requirements
- Python 3.9+ (standard library only — no
pip install). - Node 16+ (for the
npxinstaller only).
License
MIT — see LICENSE. (Copyright holder is set to smallseven; edit it if you'd
rather publish under a different name/org.)