AI Weather Assistant for Hong Kong
Repository
http://github.com/louiscklaw/mcp-hko
What This Does
This system helps AI assistants (like the one you're using now) get accurate weather information directly from the Hong Kong Observatory. Think of it as giving AI a direct line to the official weather experts.
Why It's Useful
- 🌡️ Get instant weather updates: Current temperature, humidity, and forecasts
- ⚠️ Receive weather warnings: Typhoon signals, rain alerts, and special notices
- 📱 Works in chat apps: Ask about weather naturally in any conversation
- 🕒 Always up-to-date: Automatically checks the latest official reports
How It Works (Simple Version)
- You ask your AI assistant about Hong Kong weather
- The assistant checks with the Observatory system
- You get the latest weather info in seconds
Example Questions You'll Be Able To Ask (Once Available)
- "Is it going to rain in Central this afternoon?"
- "What's the temperature in Tsim Sha Tsui right now?"
- "Are there any typhoon warnings today?"
- "Will it be sunny this weekend in Discovery Bay?"
just like picture below 👇

Behind the Scenes
The system securely connects to the Hong Kong Observatory's official data sources using special permissions. It translates technical weather data into simple answers anyone can understand.

Privacy and Reliability
- 🔒 Your data is safe: No personal information is collected
- ✅ Official sources: All data comes directly from HKO
- ⚙️ 24/7 operation: Always available when you need it
Try It Yourself
Next time you chat with an AI assistant, just ask about Hong Kong weather! The system works automatically in the background.
To add this MCP in (vscode / kilocode / claude)
{
"mcpServers": {
"mcp-hko": {
"type": "stdio",
"command": "npx",
"args": ["-y", "github:louiscklaw/mcp-hko"],
"timeout": 600
// if you know what you are doing, you can un-comment below line
// "alwaysAllow": [ "fnd", "rhrread", "flw", "warningInfo", "warnsum", "swt", "qem", "feltearthquake", "hhhot", "srs", "hlt", "hourlyrainfall", "lunardate", "mrs", "lhl", "ryes", "clmmin", "clmmaxt", "clmtemp", "ltmv"],
}
}
}
Remarks / 小記 (人腦 🧠 打出黎)
MCP 暫時缺點
MCP 可以令 LLM 直接由 internet 度攞資料,但係始終都係有D煩(bulky)。
主要原因因為 MCP instance 唔可以共用,以致唔同 MCP client 會打開唔 MCP server 去招呼佢。
每個進程都會獨立 spin 一個 process 出嚟,假使你有三四個 editor (e.g. vscode), 佢就會將同一個 process (e.g. node) spin 三四次出嚟, 越開得多,閒置process就會越多 我部 11 代 i5 開多幾個之後就熱到可以煎蛋。
解決方法唔係冇,搵到嘅都有以下呢個
mcp gateway:
https://github.com/AmoyLab/Unla
原理就係將個 MCP server,放喺個 gateway(docker) 入面打開佢, 然之後再assign一個 endpoint 畀個 mcp-server,mcp-client 要接入去果陣時就接個 endpoint 就可以
但係咁做就會將成個 solution (e.g. vscode) 變到好複雜, 當你 troubleshoot 嘅時候就多樣嘢要 troubleshoot,喱樣嘢係我唔想嘅。
幻想一下幫你打開個editor嘅時候, 你要將個 gateway 再打開先, 左攪右攪一大輪然之後你先可以用你嘅editor, 係一件幾咁麻煩嘅事?!
Prompt history
then please read
`<workspace>/mcp-hko-src/readme.md`
current i have copied the markdown file from else where and the path is `<git-repo-home>/project/mcp-hko.mdx`
while maintain the format and deep-level, please update the content of the `<git-repo-home>/project/mcp-hko.mdx` file.
the target audience is the people from outside world who doesn't know much about programming.
Please simplify and explain the project idea.
Thanks a lot !