Official MCP Service

Agent applications

In an agent application, the Large Language Model (LLM) decides whether to call an MCP service based on the conversation input.

1. Create an agent

   Go to Application Management in Alibaba Cloud Model Studio. Click Create Application. Select Agent Application > Create Now.

2. Add MCP services

   An agent can call multiple MCP services in a single response. You can add up to five MCP services at a time.

   

3. Test the MCP Server

   The agent automatically calls the appropriate MCP service to solve problems. Here are some example scenarios:

   Route planning (single MCP service)

   Make sure you have added the Amap Maps MCP service to help the agent retrieve geographic information.

   In the chat pane on the right, send this test message: “I am leaving now from Hangzhou Xiaoshan International Airport to West Lake Scenic Area in Hangzhou. Please provide three public transportation options.” The agent will call the MCP service multiple times to plan routes and estimate travel time.

   

   Step-by-step reasoning (single MCP service)

   Make sure you have added the Sequential Thinking MCP service to help the agent reason through logic problems step by step.

   In the chat pane on the right, send this test message: “Please solve this “chickens and rabbits in the same cage” problem: There are 12 heads and 32 legs in total. How many chickens and rabbits are there?” The agent will call the MCP service multiple times to complete the logical reasoning task.

   

   Temperature trends (multiple MCP services)

   Make sure you have added both the Amap Maps and QuickChart MCP services to help the agent retrieve real-time weather forecasts and plot temperature trends using line charts.

   In the chat pane on the right, send this test message: “Plot a line chart showing the temperature trend in Hangzhou for the next few days.” The agent will call the MCP services multiple times to query weather and plot the chart.

   

Workflow applications

An MCP service may offer multiple tools. In a workflow application, however, each MCP node can use only one tool. Therefore, you must manually specify the input parameters for the MCP node and pass its output parameters to the next node.

Use the maps_weather tool from the Amap Maps MCP service to quickly build a city weather query workflow.

When you use only the maps_weather tool, the workflow cannot answer questions unrelated to weather queries.

1. Create a workflow

   Go to Application Management in Alibaba Cloud Model Studio. Click Create Application. Select Workflow Application > Create Now.

   

2. Configure the start node

   Delete the two default parameters of the start node.

   In this example, you only need the default system variable query, so you can delete the other two parameters.

   

3. Extract the city name

   To parse the natural-language input into the city name that the MCP node requires, drag a Large Language Model (LLM) node from the toolbar on the left. Name it “Information Extraction”, connect it to the start node, and configure the parameters as instructed.

   Keep all other parameters at their default values.

   Parameter name	Parameter value
   Model configuration	Qwen-Max
   System Prompt	You are an expert in information extraction. Your job is to help the “weather query tool” parse the correct input parameter. --- Description of the “weather query tool”: Name: maps_weather Description: Query the weather for a specified city by city name or standard adcode. Input parameter: city [string] — city name or adcode Output parameter: result [string] — weather forecast for the city --- Your output is the input for the “weather query tool”. Output only the city name. Note: Output only one city per response.
   User Prompt	Type "/" to open the variable menu. Select “System Variables > query”.

   Note

   Before you use an MCP node, you typically need to convert natural-language input into the required input parameters. Therefore, describe the function, input, and output of the MCP service in the System Prompt to control the output format of the LLM.

   

4. Query weather information

   Use the weather query tool from the Amap Maps MCP service. It accepts a city name and returns the city’s weather information.

   Drag an MCP node from the toolbar on the left. Select the maps_weather tool as shown in the diagram. Then, confirm the node configuration.

   

   Name the MCP node “Weather Query Tool”. Then connect it to the “Information Extraction” node.

   Note

   Click the arrow button in the top-left corner of the node to expand or collapse its configuration.

   

   Click the configuration button for the maps_weather tool inside the MCP node. Set the input to “Reference: Information Extraction/result”.

   

5. Summarize weather information

   To convert the result from the MCP service into easy-to-read natural language, drag a Large Language Model (LLM) node from the toolbar on the left. Name it “Information Summary”, connect it to the “Weather Query Tool” node, and configure the parameters as instructed.

   Keep all other parameters at their default values.

   Parameter name	Parameter value
   Model configuration	Qwen-Max
   System Prompt	You are an information processing assistant. Your job is to combine user input into natural language.
   User Prompt	Type "/" to open the variable menu. Select “Weather Query Tool > result”.

   

6. Output the query result

   Connect this node to the “Information Summary” node and configure the parameters as instructed to obtain the final output.

   Parameter name	Parameter value
   Input field	Type "/" to open the variable menu. Select “Information Summary > result”.

   

7. Test the query

   Click the Test button in the upper-right corner, enter "Query Hangzhou weather", and view the workflow output.