Skip to content
An MCP server for any AI assistant

Learn a language in the conversations you already have.

langweave tracks what you can read, and your assistant writes exactly that much of its reply in your target language. Nothing to study, nothing to schedule.

Get startedRead the docs
A paragraph an assistant wrote, shown on a learner's day 1, day 30 and day 90: more of it is in the target language as the learner knows more.

El sábado por la mañana quiero ir al mercado temprano, buy pan fresco y un poco de fruta, then walk to the playa con mi hermana. Si hace buen tiempo, comeremos fuera and come home before la cena.

profile v41 · 1,206 known · 58 learning

On 土曜日の朝どようびのあさ I want to go to the 市場いちば early, buy fresh パン and some 果物くだもの, then walk to the 海うみ with my 妹いもうと. 天気がよければ、外で昼ごはんを食べて、晩ごはんの前に帰ります。

profile v41 · 1,094 known · 61 learning

Works withClaudeClaude CodeCursorGemini CLIany MCP client
How it works

Your assistant reads your profile, weaves, and logs.

fig. 1 — the profile
LANGUAGE PROFILE (format v8)KNOWN WORDS (no gloss)el edificio, el sol, la casaLEARNING WORDS (gloss on first use)- la ventana (window)- la montaña (mountain)INTRODUCE NEXT (max 2 per response)- el río (river)

Before it replies, your assistant fetches what you know: the words you can read bare, the ones you are learning, and the next to introduce.

fig. 2 — the weave
The view from the terrace is lovely — el sol setting behind la montaña (mountain), and el río (river) right below la casa.

It swaps in only words you can read, and glosses the new ones in parentheses. Everything else stays in English.

fig. 3 — the log
log_exposuresLogged 4 of 4 item(s).PROFILE CHANGED (v13) — el río: new → introduced.statses: 214 known, 31 learning, 12 introduced. 96 event(s) in the last 30 days.

Every word it uses is an exposure. Eight bare sightings and a word is known — and the next reply leaves it unglossed.

Languages

Real dictionaries, ranked by how often you'll meet each word.

  • Spanish44,481 words72,897 glosses36 grammar concepts
  • Japanese50,766 words107,222 glosses46 grammar conceptskana first, kanji when you're ready

A language is a folder. Adding one is a pull request.

Pricing

Two plans. One of them sends nothing.

Free

No card, no end date

Your profile, your words, your assistant weaves them. Nothing from your conversations ever leaves your assistant.

  • Choose a language, mark the words you know, change your settings
  • Your assistant reads your known and learning words, and your stats
  • Your profile grows only when you add words in the app

On Free, nothing from a conversation is sent to langweave or stored: your assistant sends only the language code and a bearer token.

Start free

Standard

Billed monthly · cancel any time

€4.99 / $4.99 a month, first month free. Your assistant logs what it used and introduces new words as you go.

  • Everything in Free
  • Your assistant logs the words it used, so words become known on their own
  • New words introduced as you go, on the topic you are talking about
  • Cancel any time; your words are kept

On Standard, your assistant sends the target-language words it used or wants to look up, an event kind and an optional topic hint — never your prompts or its replies — and langweave stores them as your history and the word states worked out from it.

Start your free month

Exactly what leaves your machine on each plan is on the Plans pageof the docs. See also the terms and the privacy policy.

Connect

One URL, in the assistant you already use.

Make an account and pick a language, then add the connector.

Your client

Add a remote MCP server with the URL below. Sign in when your client opens the page, approve, then just talk. Streamable HTTP, OAuth — there is no API key to paste.

claude.ai, ChatGPT, Le Chat and others take the URL in their connector settings.

  1. Open Settings → Connectors and choose Add custom connector.
  2. Name it langweave and paste the URL below.
  3. Click Connect, sign in, approve. On Claude Desktop the sign-in opens in your browser.
  1. /plugin marketplace add Jesusdiazrivero/langweave-plugins
  2. /plugin install langweave@langweave — the connector and the skill that tells the model how to weave, together.

Or without the plugin:

  1. claude mcp add --transport http langweave https://api.langweave.tech/mcp
  2. Run /mcp inside Claude Code and choose Authenticate.
  1. In ~/.cursor/mcp.json, add langweave under mcpServers with "url" set to the URL below.
  2. Sign in when Cursor opens the langweave page, approve. Check it is enabled under Customize.
  1. gemini extensions install https://github.com/Jesusdiazrivero/langweave-plugins — the server and the instructions, together.
  2. Restart Gemini CLI, then run /mcp auth langweave and sign in in the browser.
https://api.langweave.tech/mcp

Sign in once with OAuth — no API key. Your client asks for consent the first time it connects; there is nothing else to paste.

What your assistant gets
  • get_language_profile
  • check_words
  • log_exposures
  • update_item
  • stats
Get startedRead the docs