DEV Community

Yosuke Sakurai
Yosuke Sakurai

Posted on

I wired Claude to my flashcard app via MCP. Here's what that actually looks like.

#mcp #claude #ai #productivity

Automating Flashcards with Claude, Cursor, and Deckbase (via MCP)

I spend a lot of time reading papers, docs, and codebases. I used to manually copy key concepts into a flashcard app afterward. Now I do not. Claude does it while I work.

This post walks through how I connected Claude and Cursor to Deckbase using its MCP server, and what you can actually do once it is wired up.

What MCP gives you here

Model Context Protocol lets Claude call external tools mid conversation. In this case, those tools are your actual flashcard library: create a deck, add cards, generate audio, export to Anki. All without leaving your editor or chat window.

The Deckbase MCP server exposes about 30 tools over a hosted HTTP endpoint. You authenticate with a Bearer token, point your client at the URL, and Claude can read and write your decks directly.

Setup (5 minutes)

1. Get your API key

Sign up at deckbase.co, go to Settings → API Keys, and generate one.

2. Add to Claude Desktop

Edit:

~/Library/Application Support/Claude/claude_desktop_config.json

Add:

{
  "mcpServers": {
    "deckbase": {
      "url": "https://www.deckbase.co/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Restart Claude Desktop. You should see Deckbase appear in the tools list.

3. Add to Cursor

In Cursor → Settings → MCP, add:

Name: deckbase

URL: https://www.deckbase.co/api/mcp

Authorization: Bearer YOUR_API_KEY

What you can do

Create a deck and fill it from a document

Paste a block of text such as a paper abstract, a README section, or a chapter and ask Claude to card it:

Create a deck called "Transformer Architecture".
Generate 10 flashcards from this text.
One concept per card.
Front = concept.
Back = explanation.
Enter fullscreen mode Exit fullscreen mode

Claude calls create_deck, then create_cards with up to 50 cards in a single batch. Your deck is live in Deckbase before you finish reading the thread.

Add audio to any card

If you are learning a language and want native pronunciation on the back of each card:

List the voices available in Japanese, then add female audio to every card in my Japanese Vocab deck.
Enter fullscreen mode Exit fullscreen mode

Claude calls list_elevenlabs_voices with { language: "ja", gender: "female" }, picks a voice, then runs generate_audio_for_card for each card.

Generate AI images for visual memory

For anatomy, geography, or anything visual:

For each card in my Anatomy deck, generate an illustration matching the front text.
Use a clean medical diagram style.
Enter fullscreen mode Exit fullscreen mode

Claude calls list_ai_image_models, then generate_image_for_card per card with a prompt derived from the card front.

Export to Anki

If you still review in Anki and just want to use Deckbase for creation:

Export my JavaScript Concepts deck as an Anki file.
Enter fullscreen mode Exit fullscreen mode

Claude calls export_deck with export_type: "full" and returns a .apkg file you can import directly.

Normalize a messy deck

When your cards drift from the template:

Check my Python Fundamentals deck for layout drift, then normalize all cards to the current template.
Enter fullscreen mode Exit fullscreen mode

Claude first calls validate_deck_layout, then normalize_cards_to_template.

The workflow that actually changed how I work

When I am reading a technical doc in Cursor, I now do this at the end of any section I want to remember:

  1. Select the relevant text
  2. Open the AI chat panel
  3. Type: Card this.

Claude reads the selection, creates 3 to 5 cards in my current deck with clean front/back separation, and confirms what it added.

The key difference is structure. One cue, one answer, no overstuffing. The cards come out usable without editing. That was not true when using generic prompts before.

What it does not do yet

  • No automatic PDF parsing. You paste the text and Claude cards it. A folder watcher pipeline is possible via the API, just not built into MCP tools yet.
  • The hosted MCP requires a Deckbase account and API key. There is also a local stdio server (node mcp-server/index.js), but it only exposes document tools, not deck and card write operations.

Worth trying if

  • You read a lot of technical material and want a way to retain it
  • You are building an AI study workflow and want a real API instead of maintaining your own schema
  • You already live in Cursor and want your flashcard system one prompt away

The MCP docs are at:

https://deckbase.co/mcp

Top comments (0)