For months I was sending my girlfriend messages in Farsi that technically said the right words but read like a tax form. Google Translate gave me one option, no indication of whether it sounded tender or transactional, and no audio — so I had no idea if my vowel lengths were off enough to change the meaning.
I built konid to fix this for myself and anyone translating into a language they're still learning.
The core behavior: you give it a phrase you want to say, it returns three versions ordered casual to formal, with a note on what register each one occupies and how the choices differ culturally — not just lexically. For languages with tone or vowel-length sensitivity (Mandarin, Arabic, Japanese, Hindi), it plays the audio directly through your speakers using node-edge-tts, no API key required. Hearing the actual tone contour once is worth more than reading a phonetic transcription five times. The difference between a short and long vowel in Arabic, or a rising vs. falling tone in Mandarin, doesn't stick from text — it sticks from sound.
The audio isn't decorative. It's the part that changes what you remember.
Installation in Claude Code is one line:
claude mcp add konid-ai -- npx -y konid-ai
It also runs in Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. If you're on ChatGPT, Developer mode lets you add the endpoint https://konid.fly.dev/mcp directly.
Supports 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and others.
MIT licensed. https://github.com/robertnowell/konid-language-learning
Top comments (0)