Every day I was translating work emails in a language I'm still learning. I'd paste text into Google Translate, get one answer with no explanation of why, send the email, and learn nothing. Repeat. The tool got me through the task but didn't close the gap.
The thing a phrasebook — or a one-answer translation tool — can't tell you is which version of a phrase lands right in context. In Japanese, the difference between a casual and a formal register isn't cosmetic; it signals your relationship to the recipient. In Spanish, tone shifts meaning. You don't learn that from a single translation.
So I built konid: it takes anything you want to say and returns three options, casual to formal, with the register explained and a cultural nuance comparison between them. The goal was to make each translation task double as a lesson. Supports 13+ languages including Mandarin, Japanese, Korean, Spanish, French, Arabic, and Hindi. Audio pronunciation plays directly through your speakers via node-edge-tts — no external API key.
Install it once as an MCP server:
claude mcp add konid-ai -- npx -y konid-ai
Works in Claude Code, Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. Also available as a ChatGPT app via Developer mode with the endpoint https://konid.fly.dev/mcp.
The workflow I landed on: I ask for the translation, read all three options and the register notes, pick the one that fits, and send. Over time the patterns stick in a way that a single-answer tool never built. The comparison between casual and formal — seeing the actual difference, explained — is where the learning happens.
MIT licensed. github.com/robertnowell/konid-language-learning
Top comments (0)