Agent Node
The Agent node lets the model reason according to instructions and use connected MCP tools to complete tasks. It is suitable for workflows that require tool calling, external system lookup, or multi-step processing.
Quick Understanding
The Agent Node is an advanced version of the LLM Node. Besides asking AI, it can use MCP tools to access external data or tools, then answer based on the retrieved information.
Configurable Fields
| Field | Description |
|---|---|
| Label | Node name. |
| Model | Model profile used by the Agent. |
| Maximum Iterations | Maximum number of reasoning or tool-call iterations. Default is 3. |
| Enable Vision | Whether vision capability is allowed. |
| Context Variable | Selects an upstream output as context. |
| Enable Chat Memory | Whether to include recent conversation memory. |
| Memory Messages | Number of recent messages to include when memory is enabled. |
| MCP Tools | Selects MCP tools available to the Agent. |
| Messages | Configures system, user, and assistant messages. |
Screenshot pending: the current production workflow does not have an existing Agent node. After adding an Agent node in a test workflow, capture the settings panel that includes Model, Maximum Iterations, Context, Chat Memory, MCP Tools, and Messages.
MCP Tools
The Agent panel reads configured MCP servers and lists the tools exposed by each server. After selecting tools, the Agent can use them during execution.
If the MCP Tools area shows no available tools, add a server from the MCP page first and refresh tools.
Messages
The default Agent system prompt asks the model to use tools when they are helpful. You can adjust the system message to restrict tool usage scope, data handling rules, and response format.
If a Context Variable is selected, include {{#context#}} in the message so upstream context is placed into the Agent prompt.
Recommendations
- Avoid setting Maximum Iterations too high, or cost and latency may become hard to control.
- Select only the MCP tools that the task truly needs.
- For tools that modify external systems, add clear restrictions in the prompt.
- If a stable format is required, specify the output format in the system message.