I wanted to tell my girlfriend 'I missed you today' in Farsi — her first language. Google Translate gave me one answer and no indication of whether it sounded like a person or a court document. That gap is what I built konid to close.
konid is an MCP server. You install it once:
claude mcp add konid-ai -- npx -y konid-ai
and it works across Claude Code, Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. For ChatGPT users, there's a Developer-mode endpoint at https://konid.fly.dev/mcp — same server, same responses.
The angle this post is actually about: the MCP protocol means I wrote the translation logic once and it runs in four different clients without modification. That's the premise of MCP and it works here in a way that's genuinely useful, because language help isn't a one-client problem. I might want it mid-conversation in Claude Cowork, mid-commit in Claude Code, or mid-draft in ChatGPT.
What it returns
Every query comes back with three options, ordered casual to formal, with the register explained — not just labeled. If you ask how to say 'I need this done by Friday' in Japanese, you get the casual version you'd use with a close colleague, the polite version for a normal professional context, and the formal version for someone senior, with a note on what grammatical or cultural lever actually creates the difference. Audio plays directly through your speakers via node-edge-tts — no API key needed.
The three-option structure matters more than it sounds. A literal translation tool answers 'what does this say.' konid answers 'what should I actually say here, and what are my options.' Those are different questions, especially when you're learning a language and want each translation to teach you something rather than just get you through the task.
Supported languages
13+: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, Farsi, and others. MIT licensed.
Top comments (0)