Show HN: Recall – an MCP server that turns your notes into flashcards

Recall is a Show HN project that runs as an MCP server and turns notes into flashcards so you do not have to write cards by hand. You connect it to Claude or ChatGPT, point it at handwritten or typed material—PDFs, docs, or a Notion page—and it writes Q&A cards straight into a deck. You can also ask the model to research a topic and generate the same cards without first making notes. Review uses the FSRS spaced repetition algorithm. The author built it solo after repeatedly bouncing off flashcards because manual card creation was too tedious.

On the technical side, Recall sits on the Model Context Protocol path: the LLM client is the front door, and the server is what ingests sources and emits structured Q&A into a deck. Source types are explicit—PDFs, documents, Notion pages, plus free-form research prompts when you have no notes yet. The output is not a summary dump; it is card-shaped Q&A ready for review. Scheduling is handled by FSRS rather than a fixed daily queue, so interval updates follow that algorithm’s difficulty and retention logic.

For engineers and builders, the interesting move is treating flashcard creation as a tool the agent can call, not as a separate UI chore. MCP already shows up in agent stacks; wiring notes in and cards out keeps the workflow inside Claude or ChatGPT instead of copy-paste into Anki-style apps. That matters if your learning material already lives in PDFs, docs, or Notion and you want generation and review to share one pipeline. Solo-built MCP servers also signal how small tools can plug into the same client surface as larger products.

In market terms, Recall sits between note tools (Notion, PDFs, docs) and spaced-repetition review (FSRS). Hand-authored decks remain the default for serious learners, but the cost of writing every card is exactly what pushed this author away. Competing approaches either stay fully manual, ship proprietary AI card generators inside one app, or leave research and card writing as two disconnected steps. Recall’s bet is open client choice—Claude or ChatGPT—plus MCP as the integration layer rather than a closed flashcard app.

What to watch next is whether people actually run the full loop: connect an MCP client, feed real notes or a research prompt, land cards in a deck, and stick with FSRS review. Source quality will drive card quality—handwritten and typed notes, PDFs, docs, and Notion pages are supported, but noisy inputs will still produce noisy Q&A. If the research-without-notes path becomes the common use case, the product’s value shifts from “parse my notes” to “turn a topic into a reviewable deck.” The practical test is simple: does card creation stop being the reason people bounce off spaced repetition?