Translating work emails every day in a language you're still learning turns into a grinding ritual: paste into Google Translate, get one flat answer, no sense of whether it's stiff or too casual, send it, wonder later. You learn nothing and you're never confident the register was right.
konid is an MCP server that returns three options per query — casual to formal — with the register named, cultural context explained, and audio pronunciation played through your speakers via node-edge-tts (no API key required). The three-option structure is the point: you pick, and the comparison teaches you something about the language rather than just getting you through the task.
Because it's a standard MCP server, it runs the same way across whatever client you're already in. Claude Code:
claude mcp add konid-ai -- npx -y konid-ai
Cursor, VS Code Copilot, Windsurf, Zed, and JetBrains all pick it up the same way. For Claude Cowork or ChatGPT (Developer mode), point at the hosted endpoint:
https://konid.fly.dev/mcp
No separate install, no syncing configs across environments. The same tool, behaving identically, wherever you're working.
I built it because the flat-answer problem bothered me in two contexts: personal (wanting to say something that actually sounds natural to someone whose first language isn't mine) and professional (daily work emails in a language I'm still learning, where register matters and I wanted each translation to move my fluency forward, not just clear my inbox). Supports 13+ languages including Mandarin, Japanese, Korean, Spanish, Arabic, and Hindi.
MIT licensed. github.com/robertnowell/konid-language-learning
Top comments (0)