The Neural Center for Chinese Local Life Agents
English | 简体中文 | 日本語 | 한국어 | Français | Deutsch
Let Claude / DeepSeek / Qwen directly order your takeout, hail a Didi, check high-speed rail tickets, and pay utility bills.
DragonMCP is a Model Context Protocol (MCP) server designed to bridge the gap between AI Agents and local life services in Greater China (Mainland China, Hong Kong) and Asia.
It aims to solve the "last mile" problem between AI Agents and real-world services.
We have implemented the MTR (Mass Transit Railway) Query Tool as our first MVP. AI Agents can now fetch real-time train schedules directly from MTR's Open API.
Scenario:
User: "When is the next train from Admiralty to Central?"
Agent Response:
"Next Island Line train from Admiralty to Central (towards Kennedy Town):
- Arriving in: 2 min(s) (10:30:00)
- Subsequent trains: 5 min(s) (10:33:00)"
(Try it yourself by connecting DragonMCP to your MCP client!)
We are actively expanding our support for local services. Below are the currently integrated interfaces (some are mocks/placeholders for development):
| Region | Category | Service | Tool Name | Description | Status |
|---|---|---|---|---|---|
| Greater China | Travel | MTR (HK) | search_mtr_schedule |
Real-time train schedule (Island/Tsuen Wan Line) | ✅ Live |
| Amap (Gaode) | amap_search_poi |
Search for POIs (Restaurants, Hotels, etc.) | ✅ Live | ||
| Amap (Gaode) | amap_walking_direction |
Walking route planning | ✅ Live | ||
| Amap (Gaode) | amap_driving_direction |
Driving route planning (Fastest) | ✅ Live | ||
| Amap (Gaode) | amap_transit_direction |
Public transit route planning (Integrated) | ✅ Live | ||
| Weather | HK Observatory | hk_weather_current |
Current weather report in Hong Kong | ✅ Live | |
| Travel | Didi | book_taxi_didi |
Estimate price and book a ride | 🚧 Mock | |
| Payment | WeChat Pay | wechat_pay_create |
Create payment order | 🚧 Mock | |
| Alipay | alipay_pay_create |
Create payment order | 🚧 Mock | ||
| Lifestyle | Meituan | meituan_search_food |
Search for food delivery | 🚧 Mock | |
| Shopping | Taobao | taobao_search_product |
Search for products | 🚧 Mock | |
| Asia Expansion | Travel | Grab (SG/SEA) | book_ride_grab |
Estimate and book a ride | 🚧 Mock |
| Naver Maps (KR) | naver_map_search |
Search for places in Korea | 🚧 Mock | ||
| Payment | LINE Pay (JP/TW) | line_pay_request |
Request a payment | 🚧 Mock |
IMPORTANT: This project includes Mock implementations for sensitive services like payments (WeChat Pay, Alipay) and ride-hailing (Didi).
- Do NOT use real financial data or personal credentials in the current version.
- The payment tools (
wechat_pay_create,alipay_pay_create) currently return fake data for demonstration purposes only. No actual money is transferred. - When integrating real APIs in the future, ensure you follow strict security protocols (OAuth, HTTPS, Token Management).
DragonMCP acts as a middleware between AI Agents and various local service APIs.
graph TD
A[AI Agent Client] -->|MCP Protocol| B[DragonMCP Server]
B --> C[Service Router]
subgraph "Greater China Services"
C --> D["Payment (WeChat/Alipay)"]
C --> E["Travel (MTR/Amap/Didi)"]
C --> F["Lifestyle (Meituan/Taobao)"]
end
subgraph "Asia Expansion Services"
C --> G["Travel (Grab/Naver)"]
C --> H["Payment (LINE Pay/PayNow)"]
C --> I["Lifestyle (Rakuten/Kakao)"]
end
subgraph "External APIs"
D -.-> J[WeChat/Alipay API]
E -.-> K[MTR/Amap API]
G -.-> L[Grab/Naver API]
H -.-> M[LINE Pay API]
end
Architecture details are documented in this README and will be expanded in repo docs over time.
- Core Framework: Express + MCP SDK + TypeScript setup.
- Travel (MTR): Real-time schedule query for Island Line & Tsuen Wan Line.
- Travel (Amap): POI search and walking directions.
- Service Mocks: Basic structure for WeChat/Alipay/Didi/Meituan/Taobao.
- Food Delivery (Demo): Simulate ordering process (Search Shop -> Menu -> Cart).
- Basic Config: Environment variables & project structure.
- Structure Setup: Service directories for Singapore (Grab), Japan (LINE), Korea (Naver).
- Initial Mocks: Grab ride booking, LINE Pay request, Naver Map search.
- Real API Integration: Replace mocks with real APIs (Grab Developer, LINE Pay API).
- More Services: Kakao Pay (KR), Yahoo! Transit (JP), EZ-Link (SG).
- Plugin System: Allow community to contribute individual service tools.
- User Auth: Secure user token management for personal services.
- Node.js >= 18
- npm or yarn
-
Clone the repository:
git clone https://github.com/arthurpanhku/DragonMCP.git cd DragonMCP -
Install dependencies:
npm install
-
Configure environment variables:
cp .env.example .env # Edit .env (AMAP_API_KEY required for map services)
Minimum recommended variables:
AMAP_API_KEY: required for Amap search and routing tools.JWT_SECRETandENCRYPTION_KEY: required whenNODE_ENV=production.
Start the development server with SSE support:
npm run devThe server will start at http://localhost:3000.
SSE Endpoint: http://localhost:3000/mcp/sse
-
Build and start the container:
docker-compose up -d --build
-
View logs:
docker-compose logs -f
-
Stop the server:
docker-compose down
-
Build the project:
npm run build
-
Add the following to your
claude_desktop_config.json:{ "mcpServers": { "DragonMCP": { "command": "node", "args": ["/path/to/DragonMCP/dist/server.js"], "env": { "NODE_ENV": "production", "AMAP_API_KEY": "your_amap_api_key_here" } } } }(Note: Replace
/path/to/DragonMCPwith your actual absolute path)
A: Currently, only Island Line and Tsuen Wan Line are supported. Please check if the station name is spelled correctly (e.g., "Admiralty", "Central", "Mong Kok").
A: You need to register at the Amap Open Platform, create a "Web Service" application, and copy the Key to your .env file as AMAP_API_KEY.
A: No. The current payment tools are mocks. Do not use them for real transactions.
Run unit and integration tests:
npm testIf your local Node/Jest setup requires ESM flags, you can run:
NODE_OPTIONS="$NODE_OPTIONS --experimental-vm-modules" npm testWe welcome all contributions! Whether you are a developer, designer, or product thinker.
- Playwright Scripts: Simulating food delivery apps (Meituan/Ele.me) web flows.
- More MTR Lines: Adding station data for East Rail Line, Tuen Ma Line, etc.
- Real API Integration: Replacing mocks with real APIs for WeChat/Alipay/Didi.
See CONTRIBUTING.md for details.
- Anthropic: For creating the Model Context Protocol (MCP).
- MTR Corporation: For providing the Open Data API.
- Amap (Gaode): For the map and POI services.
This project is licensed under the MIT License - see the LICENSE file for details.