I translate work emails in Japanese every day. My workflow broke the same way every time: Google Translate gave me one answer, I pasted it, and three days later a colleague would gently tell me the register was off — too casual for a senior client, or stiff in a way that read as cold to a peer.
I built konid to fix that loop. It's an MCP server that returns three translations per query — casual, neutral, formal — with a register explanation and a cultural note comparing the options. Audio pronunciation plays through your speakers via node-edge-tts, no external API key required.
The angle I want to walk through here is the MCP setup, because the same server works across four clients without any changes to the tool itself.
Install once, use everywhere
For Claude Code:
claude mcp add konid-ai -- npx -y konid-ai
That's the full install. Same server also works in Cursor, VS Code Copilot, Windsurf, Zed, and JetBrains — add it through each client's MCP config pointing at the same npx invocation.
For ChatGPT (Developer mode), add a custom action with this endpoint:
https://konid.fly.dev/mcp
No separate deployment, no forked config. The MCP spec handles the protocol; the clients handle the UI.
What a query returns
Ask it "how do I say I missed you today in Japanese" and you get three options with the formality spelled out: which one you'd use with a close friend, which works in a work context, which reads as warm but appropriate with someone you respect. Cultural notes flag anything that would land wrong — not just vocabulary, but register cues native speakers use without thinking.
The tool covers 13+ languages: Mandarin, Japanese, Korean, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, and more.
Why MCP for this
I could have shipped a web app. The reason I went MCP is that language questions come up inside the tools I already have open — mid-email draft in VS Code, mid-conversation in Claude Code. A context switch to a browser tab breaks the flow the same way the wrong translation does. The query, the three options, and the audio all land inline where you're already working.
Top comments (0)