Every day I was translating work emails in a language I'm still learning. Google Translate gave me one answer, no context, no explanation of why that word and not another. I'd send the email and learn nothing. The next day, same problem.
I built konid as an MCP server. One installation, and it runs identically inside Claude Code, Cursor, Claude Cowork, and ChatGPT (via Developer mode, endpoint https://konid.fly.dev/mcp). The install for Claude Code is a single line:
claude mcp add konid-ai -- npx -y konid-ai
The reason I built it as MCP rather than a standalone app: I'm already inside these editors and tools for hours a day. Switching to a browser tab or separate app breaks whatever I was doing. The MCP protocol means konid sits where I'm already working — I ask in natural language, the response comes back in context.
What it actually returns is three options per query, ordered casual to formal, with the register explained for each. Ask how to say "I'm swamped this week, can we push the meeting?" in Japanese and you get the version you'd use with a close colleague, the version for a manager, and the version for a client — with notes on what makes each appropriate. Audio pronunciation plays directly through speakers via node-edge-tts, no external API key required.
The cultural context is the part that changed how I use translation. Literal tools give you a word-for-word mapping. konid flags where a direct translation would read as rude, overly familiar, or just odd to a native speaker. That's the difference between getting through a task and actually learning something from it.
Supports 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and more. MIT licensed.
Top comments (0)