Building AI Agents with Qwen3 and Qwen‑Agent: A Hands‑On Guide to MCP Integration

Register Qwen3 API‑Key

Log in to the Alibaba Cloud Bailei Large Model Service platform at https://bailian.console.aliyun.com/. If a prompt to enable model services appears, click “Enable Now”.

Navigate to the “My API‑Key” page at https://bailian.console.aliyun.com/?spm=a2c4g.11186623.0.0.60907980MjFErt&tab=model#/api-key and click “Create My API‑Key”.

In the API‑Key list, click “View” to obtain the key. This key is required for calling Qwen3 models.

Quick start with Qwen‑Agent

What is Qwen‑Agent?

Qwen‑Agent is an open‑source agent development framework released by Alibaba’s Tongyi team. It is built on Qwen large language models and provides tool calling, planning, memory, long‑text retrieval‑augmented generation (RAG), and UI interaction capabilities.

Build a multi‑turn dialogue bot

Install the required packages in an Anaconda environment:

conda create -n qwen_agent python=3.12
pip install -U "qwen-agent[rag,code_interpreter,gui,mcp]"
pip install uv

Import the core classes:

from qwen_agent.agents import Assistant
from qwen_agent.utils.output_beautify import typewriter_print

Configure the model service (replace 'YOUR_API_KEY' with the key obtained above):

llm_cfg = {
    'model': 'qwen3-235b-a22b',
    'model_server': 'dashscope',
    'api_key': 'YOUR_API_KEY',
    'generate_cfg': {
        'top_p': 0.8
    }
}

Create the assistant object:

bot = Assistant(
    llm=llm_cfg,
    system_message='You are a helpful assistant.',
    name='Smart Assistant'
)

Run a conversation loop:

messages = []
while True:
    query = input('
User request (type "quit" to exit): ')
    if query == 'quit':
        break
    messages.append({'role': 'user', 'content': query})
    print('AI reply:')
    response_plain_text = ''
    for response in bot.run(messages=messages):
        response_plain_text = typewriter_print(response, response_plain_text)
    messages.extend(response)

Integrate MCP tools

Overview

Qwen‑Agent can connect to MCP services using the stdio development mode, where the MCP server runs as a subprocess and the agent communicates with it via standard input/output.

Connect to mcp-server-sqlite

Define the MCP configuration and create the assistant:

def init_agent_service():
    llm_cfg = {
        'model': 'qwen3-235b-a22b',
        'model_server': 'dashscope',
        'api_key': 'YOUR_API_KEY',
        'generate_cfg': {'top_p': 0.8}
    }
    tools = [{
        "mcpServers": {
            "sqlite": {
                "command": "uvx",
                "args": ["mcp-server-sqlite", "--db-path", "test.db"]
            }
        }
    }]
    bot = Assistant(
        llm=llm_cfg,
        name='Database Administrator',
        description='You can perform CRUD operations on a local database.',
        system_message='You act as a database assistant.',
        function_list=tools,
    )
    return bot

Run a query that creates a students table and inserts a record:

def run_query(query=None):
    bot = init_agent_service()
    messages = []
    messages.append({'role': 'user', 'content': [{'text': query}]})
    previous_text = ''
    print('Database Administrator: ', end='')
    for response in bot.run(messages):
        previous_text = typewriter_print(response, previous_text)

if __name__ == '__main__':
    query = "帮我创建一个学生表,表名是students,包含id, name, age, gender, score字段,然后插入一条数据,id为1,name为张三,age为20,gender为男,score为95"
    run_query(query)

The agent automatically installs missing dependencies, then calls the MCP functions sqlite-create_table and sqlite-write_query to create the table and insert the data. After execution, a test.db file appears in the working directory.

Verify the result with the standard sqlite3 tool:

import sqlite3
conn = sqlite3.connect('test.db')
cursor = conn.cursor()
# List tables
cursor.execute("SELECT name FROM sqlite_master WHERE type='table';")
print('Tables:', cursor.fetchall())
# Query first table if it exists
tables = cursor.fetchall()
if tables:
    table_name = 'students'
    cursor.execute(f"SELECT * FROM {table_name};")
    print(f'Data in {table_name}:', cursor.fetchall())
else:
    print('No tables found.')
conn.close()

The output confirms that the students table was created and the record was inserted.