What this is for
WordByWord.io provides an MCP server, which is an endpoint that AI assistants can connect to so they can interact with your collections and terms. Once connected, the assistant can create collections, add words you discuss, fix translations, and report your learning progress.
A common use case is reading text with an assistant. If you paste a Spanish job offer and ask for a translation, you can then tell the assistant to save the words you did not know. The assistant will write the meaning in your language, extract an example sentence from the text, record the source, and provide a link to your collection. The words are saved to your account and scheduled for review.
MCP (Model Context Protocol) is an open standard. The endpoint works with any client that supports remote MCP servers, not only the ones listed below.
The address
https://api.wordbyword.io/mcp
In ChatGPT you never need this URL — the app in the directory carries it. You enter it by hand only in clients that add MCP servers manually, such as Claude. The client then handles signing in, permissions, and fetching the available actions.
You need a WordByWord.io account. The MCP server is available on all plans without a separate subscription. On the assistant's side the rules are its own: check your client's documentation for which plans may connect apps and custom connectors.
Connecting ChatGPT
WordByWord is published in the ChatGPT app directory, so there is nothing to configure by hand: no developer mode and no server URL.
- Open the app directory in ChatGPT: click + in the message box and choose Apps, or go to Settings → Apps & Connectors.
- Find WordByWord and click Connect.
- Sign in to WordByWord.io in the window that opens and confirm access.
The app then appears in the apps list of any conversation. In workspace accounts an administrator may have to allow the app first.
Interface menus update frequently. If the wording differs, look for "apps" or "connectors" in the settings.
Connecting Claude
- Open your Claude settings and navigate to the Connectors section. Depending on your version, this is located under Settings or Customize. For Team and Enterprise plans, an owner must first add the connector in the organization settings.
- Click Add custom connector.
- Paste
https://api.wordbyword.io/mcpas the server URL. Leave the advanced OAuth fields empty, as the server provides all necessary details to the client. - Click Add, then Connect. A WordByWord.io page will open for you to sign in and confirm access.
The WordByWord tools will now appear in Claude's tool list and are ready to use in your conversations.
Other clients
Any client supporting the remote MCP protocol will work. For example, in Claude Code, you can connect using a single command:
claude mcp add --transport http wordbyword https://api.wordbyword.io/mcp
Signing in and granting access
The connector does not ask for your WordByWord.io password. Adding the connector initiates a standard OAuth flow. The client redirects you to our sign-in page, where you log in with your existing account. A consent screen then displays which app is requesting access. Access is granted only after your approval.
If you reconnect the same assistant later, the consent screen will not appear again because your previous permission remains active. To revoke access, remove the connector in the assistant's settings. This invalidates its token.
The assistant can view only your collections and terms, matching what you see on the website. It cannot access other users' data, your account settings, your subscription, or your payment details.
What the assistant can do
There are twelve available actions. Clients categorize them by their potential impact. Read-only actions typically run automatically, while actions that delete or overwrite data will prompt you for confirmation.
| Action | What it does | Type |
|---|---|---|
list_collections | Lists your collections with term counts and progress | read-only |
get_vocabulary | Reads terms of one collection or of the whole vocabulary | read-only |
search | Searches your vocabulary by collection name, term or meaning | read-only |
fetch | Opens a search result in full | read-only |
get_learning_stats | Current and longest streak, term counts, status distribution | read-only |
get_languages | Lists supported languages and their codes | read-only |
create_collection | Creates a new private collection | adds data |
add_terms | Adds terms with meanings, examples and the source they came from | adds data |
change_terms_status | Changes the learning status of terms, e.g. marks them known | changes data |
update_terms | Fixes the word, the translation or the example of existing terms | overwrites data |
delete_terms | Deletes terms from a collection | deletes data |
delete_collection | Deletes a collection with all its terms | deletes data |
request_feature | Sends the team a note about a capability that is missing | sends a message |
The learning statuses match the five used in the app: new, recognize, familiar, learned, and known. The assistant can set these manually, though they typically update automatically as you study.
Deletions are permanent and cannot be undone. Always verify what the assistant intends to delete before approving the action.
What to ask for
You can use natural language for your requests; you do not need to reference the specific action names.
- "Save the words I did not know from this article into my Spanish collection."
- "Create a collection for German work vocabulary and put these ten words in it."
- "Which of my Italian words are still marked new?"
- "Fix the translation of Vorfahrt — it should be 'right of way', not 'priority'."
- "How many days is my streak?"
A single request often translates into multiple actions. For example, the assistant might first search for the collection and then add the terms. If the words originate from a specific text or webpage, ask the assistant to record the source so it appears in your collection.