own-words

io.github.alexslkgitPublished by GitHub account alexslkgitv1.0.0Updated Oct 24, 2026

Turn notes into a study deck: answer each card in your own words, then compare.

VerifiedStreamable HTTPWeb executableProductivity & WorkflowKnowledge & Memory

Overview

AI-generated overview

Remote MCP server that validates a study-deck JSON object and returns a shareable own-words link for answering flashcards in your own words.

What you can do
  • Validate a study-deck JSON object and return a shareable own-words link with the card count.
  • The deck itself travels inside the link, so the assistant can hand over a ready-to-open deck instead of a file.
When to use it
Use it when you want an assistant to turn a topic or notes into a self-testing study deck you can open in a browser and work through card by card. Useful for short decks you intend to share as a link.
Requirements
A remote MCP endpoint; the public instance is added in claude.ai under Settings > Connectors > Add custom connector with no sign-in. Opening a deck link needs a modern browser. No account or API key is required.
Before you install
The whole deck is compressed into the URL after the fragment, so long decks make very long links; the README suggests keeping long decks as files. The server keeps no state and stores nothing, and deck progress lives only in the browser's local storage, so clearing site data loses it.

Installation

In SourceWeft

  1. Open own-words in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.alexslkgit%2Fown-words/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "own-words": {
      "type": "http",
      "url": "https://own-words-mcp.alexslkmain.workers.dev/mcp"
    }
  }
}

README

own-words

own-words turns a topic into a deck of question cards. Each card asks a question, you type your answer, and only then does the prepared answer appear; you mark it "I knew it" or "see again" and go on.

I have ADHD. Reading a prepared answer first always felt like understanding, and then the next real question found out that it was not. Typing my own answer first is what stuck, so the deck now makes me do it before it shows me anything.

New here? Try the 2-minute tutorial.

[A card with the prepared answer open above my own words]

Try it, nothing to install

https://alexslkgit.github.io/own-words/

Two decks are on that page: "How own-words works", 8 cards, and "Claude Code in 30 questions", 30 cards. Open one, answer three or four cards, and you have seen the whole thing.

Make your own deck

No account for any of these.

Paste deck JSON into the box on the landing page. A fenced reply or a line of chat around the JSON is fine; the box finds the deck inside it.

Press "Copy the prompt" on the same page, hand the prompt to any assistant along with your topic, and paste back what it replies with.

Install the Claude Code plugin:

claude plugin marketplace add alexslkgit/own-wordsclaude plugin install own-words@own-words

Then run /own-words with a topic or a file. It writes <topic-slug>.deck.json in the current directory, checks it, and prints the link that opens it.

A deck is one JSON file: a key, a title, and blocks of cards, each card a question and an answer. The two decks above are in decks/ in exactly that form, with nothing in them the format does not give you. The file is also how a deck moves to another device: the link for the 30-card demo is about 17 KB because the whole deck travels inside it, so keep long decks as files and share the link for short ones.

A card may also carry a weight, "w": 1 to 100. When any card in a deck has one, three stops in the header show the whole deck, only cards weighted 60 and up, or only 85 and up, so a short session covers what matters most. A card with no weight is always shown.

When an answer is too hard to read

Once the answer is shown, tap the words you do not follow, or select a phrase, then press "Simpler, not longer". The page builds a request whose word budget is the answer's own length; send it to Claude or any chat and paste the reply back. The page refuses a reply with more words than the budget, so the card gets clearer without growing. With the plugin, Claude Code also writes the simpler version into your .deck.json beside the original answer, which it never changes. For example, with "existential" and "witness table" marked:

Before, 30 words: some names one concrete type that the compiler knows but hides, while any is an existential box resolved at runtime through a witness table, so its calls cannot be specialized.

After, 26 words: some: one real type the compiler knows, just unnamed. any: a box (fits any matching type) opened at run time, so its calls cannot be specialized.

Where the deck lives

The link carries the whole deck inside the URL, after #d=, compressed with lz-string. A URL fragment is never sent in the HTTP request, so a deck link opens without any server ever holding the deck. deck.html renders whatever deck it is handed and knows nothing else.

Every deck you open is kept in this browser's localStorage, under ownwords:decks:<key> and ownwords:index, together with your answers and marks; the landing lists them under "My decks". There is no account and nothing you write is stored on a server. Clearing site data for the page drops the decks and the progress, and "My decks" empties; the .deck.json file or the link brings a deck back, the progress does not come back.

The one time your text moves is the copy button on a card page. It builds a plain text form of the deck: every card you touched, with the question, your own wording, and your mark, then the numbers of the cards you left alone. Nothing is copied until you press it. Pasting that form back into the chat that wrote the deck is how you ask for harder cards.

Plain HTML and JavaScript, no build step, one vendored dependency (lz-string). The pages ask Google Fonts for two typefaces; block that and everything still works in a system font, including from a file:// path.

connector/ is a small remote MCP server so claude.ai can hand you a deck link directly. A public instance runs at https://own-words-mcp.alexslkmain.workers.dev/mcp: in claude.ai open Settings > Connectors > Add custom connector, paste that URL and pick No sign-in. It keeps no state and stores nothing, the deck goes into the link. To run your own, see connector/README.md.

Uninstall

claude plugin uninstall own-words@own-wordsclaude plugin marketplace remove own-words

The site needs nothing installed, so closing the tab is enough.

Requirements

A modern browser, for the site and for any deck link. The plugin needs Claude Code.

This is one person's tool at version 0.3.0. I use it daily on my own decks; the English strings in it are new, so if a line reads oddly, that is why.

If you tried it on something you actually needed to learn, did writing your own answer first change what you remembered a week later? Open an issue and tell me: https://github.com/alexslkgit/own-words/issues

MIT, see LICENSE.

Source: README.md at commit d3a11b6

Tools

1

create_deck

Validate a study-deck JSON object and return the shareable own-words link with the card count.

Parametersdeckrequired

Version history

1
  1. v1.0.0LatestOct 15, 2026

Score

100% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts