Build a Chat & Poetry Creation Agent with Qwen3 and Alipay MCP Using Tongyi Lingma

Prerequisites

Tongyi Lingma version 2.5 or higher installed

Ollama installed

Alipay MCP configuration completed (environment variables AP_APP_ID, AP_APP_KEY, AP_PUB_KEY)

Step 1 Generate the agent code

In an IDE (e.g., PyCharm) open the Tongyi Lingma panel, select the “Intelligent Conversation” mode, choose the qwen3‑coder model and provide a prompt that describes the required capabilities: daily chat, poetry creation, and payment request. The model references the Qwen‑Agent implementation, adapts it to the Chainlit framework and creates a Python project (e.g., tongyi_lm_demo_06).

Reference: https://github.com/QwenLM/Qwen-Agent
Prompt (Chinese):
“你是一个智能陪伴和创作助手，可以进行日常聊天和诗词创作，会使用温暖热情的语气与用户交流。实现的功能包括：
1、日常聊天：回答问题、进行对话；
2、诗词创作：根据用户给定的主题创作诗词，并提供简短解释。”

Step 2 Run and test

Activate the virtual environment and launch Chainlit: .venv/bin/chainlit run app.py -w The web UI appears; interact with the agent to verify chat and poetry functions.

Step 3 Add Alipay MCP payment

Extend the system prompt to include a payment request after delivering poetry. The agent should call the tool create-web-page-alipay-payment to generate a payment link (amount 0.01‑0.03 CNY) and a random order ID based on the current timestamp. After the user confirms payment, the agent calls query-alipay-payment to check the status and respond accordingly.

Prompt addition (Chinese):
“3. 请求打赏：在提供诗词创作内容后，调用工具 create-web-page-alipay-payment 生成网页支付链接，金额在 0.01‑0.03 元之间，订单号根据当前时间戳随机生成，订单标题根据上下文生成，并向用户展示订单号、金额和支付链接。
4. 表达感谢：如果用户已打赏或已支付，调用工具 query-alipay-payment 查询订单状态；若已完成支付，输出至少三句夸奖；若未完成，提醒用户完成支付。
注意：每次都必须真实调用工具，禁止自行 mock。”

MCP server configuration (JSON‑like snippet):

{
  "alipay-server": {
    "command": "npx",
    "args": ["-y", "@alipay/mcp-server-alipay"],
    "env": {
      "AP_APP_ID": "${AP_APP_ID}",
      "AP_APP_KEY": "${AP_APP_KEY}",
      "AP_PUB_KEY": "${AP_PUB_KEY}"
    }
  }
}

Troubleshooting

If the agent prints duplicated output, modify typewriter_print in output_beautify.py (see https://github.com/QwenLM/Qwen-Agent/blob/main/qwen_agent/utils/output_beautify.py) to emulate printer‑style non‑repeating text.

Appending /no_think to message.content suppresses the model’s internal “thinking” and speeds up responses.

When tool invocation format is incorrect, follow the example in assistant_mcp_sqlite_bot.py (see https://github.com/QwenLM/Qwen-Agent/blob/main/examples/assistant_mcp_sqlite_bot.py).

Ensure the tool’s return value is printed (use the adjusted typewriter_print) so the payment link appears complete.

Generate unique order IDs by adding a timestamp‑based random string utility.

Optional enhancements

After a successful payment, call query-alipay-payment to confirm the transaction and display a thank‑you message.

Reference resources

Qwen‑Agent repository: https://github.com/QwenLM/Qwen-Agent

Chainlit documentation (for running the UI)

Alipay Open API reference: https://opendocs.alipay.com/open/0h3he5?pathHash=87d45