How can I pass output from one AI Agent tool to another in n8n workflows?

Below is some clarification on how the AI Agent node handles tool execution and data flow.

How the Tools Agent Works

The AI Agent node (using the Tools Agent) is designed so that it uses external tools (which can include sub-workflows) to do specific actions based on the agent’s decisions. In this setup, the agent makes individual tool calls, and each call is treated as an independent function invocation rather than forming a strict pipeline where each tool’s output automatically becomes the input of the next tool.

Chaining Data Between Tools

Currently, there isn’t a built-in mechanism to automatically pass the processed data from the output of one tool directly as input to the next tool in the agent’s flow. In other words, while you can attach several sub-workflows (tools) to the AI Agent node, the agent itself does not automatically chain the outputs from one tool to become the parameters for the next.

Workarounds and Recommendations

If you would like to simulate a chain-based workflow (or create a “flexible, agent-at-the-center workflow”) you might consider these approaches:

• Designing Custom Sub-workflows:
  Instead of relying on the agent to automatically route the result from tool A to tool B, design your sub-workflows to accept parameters that you set explicitly. You might have a sub-workflow that, upon completion, stores its output in a way that the next sub-workflow (which is invoked as a separate tool) can access it. This might involve using persistent storage or global variables within n8n.

• Using the $fromAI() Function:
  If your use case allows some dynamic parameter population, you can use the $fromAI() function. This function enables an AI agent to determine parameters at runtime. While it doesn’t link outputs directly, it does allow your agent to specify input parameters for a tool based on previously gathered information.

• Returning Intermediate Steps:
  The agent has options (such as the “Return Intermediate Steps” option) that let you inspect the chain-of-thought or responses from previous tool calls—but this feature is for debugging or logging rather than automatically connecting outputs as inputs.

In Summary

Directly chaining the processed data from one tool to the next isn’t supported by default with the Tools Agent node. Each tool call remains an isolated action decided on by the agent. To achieve the desired chaining effect, you’ll need to architect your workflows so that intermediary outputs are persisted and then manually picked up by subsequent tool calls (for example, via custom code or using n8n’s dynamic parameters features).

This design keeps the agent flexible but places state management (or data propagation) within your control as you build the solution.