When a workflow needs to invoke external capabilities—such as time zone conversion, web search, or image generation—integrate third-party services—like Alipay payment or Amap—and extend large model functionality—such as code execution or speech synthesis—the MCP node connects to service tools via the standardized MCP protocol, enabling workflows to retrieve real-time data and perform specific actions.
Usage Example
Below is a typical workflow that uses an MCP node to fetch weather information for Hangzhou. It calls the maps_weather tool from the Amap MCP service and passes the result to a Large Language Model (LLM) node for summarization.
Prerequisites
Before integrating an MCP node, you must first enable or create an MCP service.
- Enable an official MCP service: Go to the MCP Plaza, select the desired MCP service card, and click Enable Now.
- Create a custom MCP service: On the MCP Management page, click Create MCP Service. For details, see [documentation].
Adding the Node
- Click the + button at the end of any node on the canvas.
- In the node selection panel, choose MCP.
-
In the MCP service selection dialog:
- MCP Plaza: Select an official MCP service pre-configured by Alibaba Cloud Bailian.
- Custom MCP: Select a deployed custom MCP service.
- Expand the service card and choose the specific tool you want to invoke.
- Click OK; the node is automatically added to the canvas.
Core Configuration
Tool Configuration
After selecting an MCP service, specify the exact tool to invoke. Each MCP service may expose multiple tools, but each MCP node can only call one tool.
Click the toggle button to reselect the MCP service and tool.
Input Configuration
Once a tool is selected, its defined input parameters are automatically loaded into the input section.
| Configuration Item | Description |
|---|---|
| Variable Name | Defined by the tool and cannot be modified. |
| Reference Method | Choose how to provide the variable value: Reference (from upstream nodes) or Input (static value). |
| Value | Enter a static value or reference a variable from upstream nodes. Typing / in the input field triggers the variable selector. |
Node Output
The output variables of an MCP node follow this structure:
| Variable Name | Type | Description |
|---|---|---|
result | Object | The tool execution result. |
result.content | Array<Object> | Returned content containing the actual output data from the tool. |
result.isError | Boolean | Indicates whether an error occurred: true means execution failed. |
Common Use Cases
| Paired Node | Description |
|---|---|
| Large Language Model (LLM) Node | Use the MCP tool’s return result as input to the LLM node for content generation or analysis. |
| Conditional Branch Node | Use result.isError to determine whether the tool executed successfully and implement conditional logic. |
| Loop Node | Invoke the MCP tool in batch to process multiple data items. |
Frequently Asked Questions
What should I do if the MCP node fails with a “tool invocation error”?
This may occur due to incorrect input parameter formatting or missing required fields.
- Verify all required input parameters are provided.
- Confirm parameter formats comply with tool requirements (e.g., time zone format must be
Asia/Shanghai). - Visit the MCP Management page to review the tool’s parameter documentation.