Every day I was translating work emails in Japanese — same tool, same pattern: paste, copy, send. Six months in, I could not write a single sentence from memory. The tool had done the job and left me exactly where I started.
konid is built around the loop I was missing. You ask it how to say something, and instead of one answer, you get three — ordered casual to formal — each with a register note explaining who uses it and when, plus a cultural context comparison across the options. The goal is that after ten queries, you understand something you didn't understand before.
Here's what the install looks like in Claude Code:
claude mcp add konid-ai -- npx -y konid-ai
And in Cursor, VS Code Copilot, Windsurf, Zed, and JetBrains it runs as an MCP server via the same package. If you're on ChatGPT with Developer mode enabled, the endpoint is https://konid.fly.dev/mcp.
The thing that changed how I used it: audio pronunciation plays directly through your speakers via node-edge-tts — no external API key required. So the loop is: see three options, read the register notes, hear how each one sounds, pick one. That's closer to how an actual teacher would handle the question than any translation widget I've used.
Supports 13+ languages including Japanese, Mandarin, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, and Hindi. MIT licensed.
github.com/robertnowell/konid-language-learning
Top comments (0)