I wanted to say something personal to someone in their first language — not just accurate, but natural. Every translation tool I tried gave me one answer, no explanation of whether it was casual or stiff, no way to hear what it actually sounded like.
The problem with reading a transliteration is that tone contour and vowel length don't survive text. In Mandarin, the difference between mā (mother) and mǎ (horse) is a pitch curve, not a spelling. In Arabic, a long vowel held half a beat too short shifts the word. You can memorize the romanization and still produce something unrecognizable.
So I built konid: it returns 3 options per query, ordered casual to formal, with the register difference explained and cultural context for each. But the part that changed how I actually remember phrases is the audio — it plays pronunciation directly through your speakers via node-edge-tts, no external API key, no copy-paste into a separate tab.
Hearing the same phrase at three register levels back-to-back is different from reading them. The formal version of a Spanish phrase often has a different rhythm than the casual one. The polite Japanese option lands the honorific suffix in a specific position that changes the prosody. Listening to the contrast between options is what locks in which one you actually want.
konid runs as an MCP server, so I use it inside Claude Code while I'm already working:
claude mcp add konid-ai -- npx -y konid-ai
Also works in Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. If you use ChatGPT, there's a Developer mode install via https://konid.fly.dev/mcp.
13+ languages including Mandarin, Japanese, Korean, Arabic, Hindi, Russian, and the major European ones.