Every work email I translate in Japanese is a missed lesson

For the past year I've been translating work emails in Japanese — a language I'm still learning — using Google Translate and DeepL. Both tools gave me one answer. I shipped it, moved on, learned nothing.

The problem isn't accuracy. It's that a single translation strips out the decision you actually need to make: how formal is this register, and why? Japanese business email lives and dies on that distinction. "Please review the attached" in casual Japanese to a senior colleague reads like a mistake, not just an imprecise word choice.

So I built konid. For any phrase you want to say, it returns three options ordered casual to formal, with the register explained and a nuance comparison between them. Audio pronunciation plays directly through your speakers via node-edge-tts — no external API key.

The loop I was missing:

# In Claude Code:
claude mcp add konid-ai -- npx -y konid-ai

# Then ask:
"How do I say 'please review the attached document' in Japanese?"

# Returns:
# 1. [Casual] 添付を見てください — used with close colleagues or peers
# 2. [Neutral] 添付をご確認ください — standard business register
# 3. [Formal] 添付をご査収くださいますようお願い申し上げます — for senior contacts or unfamiliar clients
# + cultural note on when each would feel off

Every translation is now a decision point: I pick the right option, read why, hear it pronounced, and remember the distinction. After a few weeks of work emails this way, I've actually stopped reaching for the tool for phrases I've asked about before.

konid works as an MCP server with Claude Code, Cursor, VS Code Copilot, Windsurf, Zed, JetBrains, and Claude Cowork. Also installs as a ChatGPT app via Developer mode at the endpoint https://konid.fly.dev/mcp. Supports 13+ languages including Mandarin, Korean, French, Arabic, Hindi, and others.

MIT licensed. https://github.com/robertnowell/konid-language-learning