I was translating a work email in Japanese and realized I'd been using the same mid-register phrasing for everything — emails to my manager, to a vendor, to a teammate I've known for years. Google Translate gave me one answer each time and never told me if it was too stiff, too casual, or just technically correct but socially weird. konid returns three options per query, ordered casual to formal, with a brief explanation of what makes each register appropriate and where it would land wrong. For Japanese, that difference between options isn't cosmetic — keigo mistakes in a business email read as either arrogant or absurdly stiff depending on which direction you miss. The tool also plays audio pronunciation directly through your speakers via node-edge-tts, no API key required. For tonal languages like Mandarin, or for Arabic where written vowels get dropped, hearing the output matters more than reading it. The MCP setup I built konid as an MCP server so it works inside the environments I'm already in. One install command: claude mcp add konid-ai -- npx -y konid-ai That covers Claude Code, Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. If you're in ChatGPT with Developer mode on, you can add it as a custom app pointing at https://konid.fly.dev/mcp — same underlying server, no separate configuration. The reason I went MCP rather than a standalone app: I want the translation layer inside the tool I'm writing in, not a separate tab I have to context-switch to. When I'm drafting a client email in Cursor, I can query konid without leaving the buffer, get all three register options with notes, pick the right one, and keep writing. What it covers 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and others. Each response includes cultural context beyond the literal words — the kind of note a fluent speaker would give you if you asked "does this sound natural?" MIT licensed. github.com/robertnowell/konid-language-learning