MCP nodePage 1

What is MCP?

MCP (Model Context Protocol) is an open-source standard for connecting AI applications to external systems. Using MCP, AI applications like Alli, ChatGPT, and Claude can connect to data sources (local files, databases, etc.), tools such as search engines and calculators, and specialized prompt-based workflows to retrieve key information and perform tasks. For a more detailed explanation of MCP, here please refer to.

How is MCP used in Alli?

Alli uses MCP when creating LLM-based workflows. Inside the app, you can connect the LLM with server-tool collections registered on the Alli MCP Hub, allowing the LLM to select and use appropriate MCP tools according to the user’s request to build workflows. For detailed guidance on this process, Agent-style apps please refer to the user guide.

What is an MCP node?

Meanwhile, as the MCP ecosystem has grown and the standard has become established, useful MCP tools have rapidly increased. Along with this, there emerged a need to use MCP tools directly like APIs—without going through an LLM—to build richer automated workflows. To address this, Alli introduces MCP nodes within interactive apps. By using MCP nodes, you can directly call pre-integrated MCP tools to construct more sophisticated automation logic and workflows.

How to use

1. Enter the interactive app and select the MCP integration node.

1. In the node editor, press the "Add MCP Server" button to enter the screen below.

1. In area 1 select a server, in area 2 choose the tool to use. In area 3 you can check the description of the MCP tool and required parameters.

2. Choose the desired tool and press the "Add" button to return to the node editor. Unlike non-deterministic agents where which tool will be called is uncertain, this node calls the API only once in a fixed manner, so you can select only one tool.

1. After selecting a tool and returning to the editor, you can see three main areas as shown above.

   1. Area 1 is the MCP server area. Click the clip icon to enter fields requested by the server, such as authentication information (API Key, OAuth authentication, etc.), to link the server.

      1. You can delete, change, or reconfigure the selected server within the node.

   2. Area 2 is the MCP tool area. Here you can change the tool, enter parameters needed when calling the tool, and switch the dropdown icon (@, Tt) to the left of a field to input a variable or plain text.

   3. Area 3 is where you assign the received response to variables. Under Received Data > click the "Add Variable" button to decide which variable will store the call result. Since all responses in the MCP specification come in JSON format, you need to parse that JSON and put it into variables. Detailed guidance for this process is provided under the "?" icon. See the image below.

      1. If an MCP call fails, the error message is stored in the @MCP_ERROR_MESSAGE variable (string). You can use it to check the reason for failure.

Guidance shown when clicking the "?" icon - Basic

Guidance shown when clicking the "?" icon - Example

1. Once all settings are complete, press the "Apply" button and the automation using MCP is ready.

Creating a simple example app

We will integrate an MCP that returns the current time for the given timezone and output the results for success and failure. The execution result is as follows.

The node flow is as follows.

Detailed node settings are as follows.