My girlfriend's first language is Farsi. I wanted to send her something tender — not 'I missed you' copy-pasted from a phrasebook, but something that sounded like I'd thought about how she actually speaks. Literal translation tools don't tell you whether the phrase you're using is intimate or formal, affectionate or clinical. They give you one answer and move on.
So I built konid: it returns three translations per query, ordered casual to formal, with the register explained and cultural context on why the options differ.
Here's what a query looks like installed in Claude Code:
claude mcp add konid-ai -- npx -y konid-ai
Then in Claude Code, Cursor, or any MCP-compatible editor, you ask:
How do I say 'I missed you today' in Farsi?
And you get back three options. The first is the kind of thing you'd say to a close friend or partner — the register note explains when it fits and why. The third is more formal, useful if you're writing to someone's family. The middle option sits between them. Each one has audio pronunciation played directly through your speakers via node-edge-tts, no external API key required.
The same pattern holds for work: I translate emails in a language I'm still learning, and I want each translation to teach me something about register — not just get the task done. With a single flat answer, you never build intuition about formality. With three labeled options and nuance notes, you start to.
konid supports 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and more. It also installs as a ChatGPT app via Developer mode at https://konid.fly.dev/mcp.
MIT licensed. Source at github.com/robertnowell/konid-language-learning.
Top comments (0)