Literal translation gets you through the task. It doesn't get you through the moment.
I built konid after a specific failure: I wanted to say something tender to someone in her first language, not in mine, and every translation tool gave me the neutral-register version. Technically correct. Emotionally flat. The kind of phrase that communicates competence, not care.
The fix wasn't a better translation. It was acknowledging that most phrases worth saying have at least three versions — one that fits a close friend at 11pm, one that fits a colleague in a polite email, one that fits a long relationship where nuance carries weight. konid returns all three, casual to formal, with the register difference explained and a cultural note on what each choice signals.
One real detail that mattered in building this: pronunciation. Most translation tools give you text and leave you to guess at the sounds. konid plays audio directly through your speakers using node-edge-tts — no API key, no clipboard-to-browser workflow. You hear the phrase in the register you chose, immediately.
The MCP server setup takes one command in Claude Code:
claude mcp add konid-ai -- npx -y konid-ai
Same integration works in Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. If you're on ChatGPT, Developer mode with endpoint https://konid.fly.dev/mcp connects it there too.
Covers 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and more.
The name — konid (کنید) — is Farsi for 'do.' Take action seemed like the right framing for a tool that's supposed to get you to actually say the thing.
MIT licensed. https://github.com/robertnowell/konid-language-learning
Top comments (0)