Currently the tools an agent can call are limited to single nodes. If I want it to call a workflow, I have to make that node a call workflow tool, and then host my workflow either in external workflows, or, if I want multiple workflows in a single file that the tool can call, use a switch node. It’s a time consuming an unintuitive way of building agents. It would be far easier if in the workflow editor you could attach workflows directly to tool nodes.
Thanks a lot for that suggestion. I agree, would make some things easier. The very first version I implemented even worked that way (was never publicly released and just exist for a few hours). But then changed it as I did fear that it would be quite confusing having kind of multiple “layers” combined into a single one.
People could then for example start to have a Merge Node which gets data from the main workflow and the tool one.
Or workflows would get even harder to debug because this tool-workflow would potentially be called multiple times. Displaying then multiple executions (main one + n tool ones ) in one view would be painful and confusing.
For reason like that (there were even more that I do not remember right now anymore) did I decide that it is probably not best.
Thank you for the reply. I definitely understand the difficulty of making it intuitive. My only other comment, though, is that more and more customers are going to be coming to n8n primarily for the Agentic capabilities, and less for the workflows. Catering to Agentic AI creators over Workflow creators might be a necessary choice.
There may be a misunderstanding. I did not mean we did not implement it because we thought the “workflow creator” experience was more important. We did not implement it because it would generally be probably more confusing and make debugging horrible without a bigger rework.
And to make it clear. Also, I’m not saying it would not be possible; we would never do it, and so on. It is just something that would not be simple or fast and would probably require more significant resources as a lot of research would have to be done as well.
Thank you so much for the response. That makes sense
Hi, I thinks this feature request would deserve a new review, as agentic frameworks has evolved a lot since then.
LangChain v1 introduced the agent loop as the main primitive in October 2025, Mastra is now a mature solution in the JS ecosystem and does have a workflows-as-tools mechanism, Vercel AI SDK does have the tool loop agent etc.
So people are more and more used to setting up agents with tools, and are aware of how tools work.
I think the counter-points made here are 100% valid but could still be addressed in some ways:
When using code, the distinction between a tool and “just a function” (so a node in n8n ecosystem) is more about the wrapper around the function: a tool is expected to have a name, a description, maybe labels, permissions rules, and should output readable errors rather than crashing, while a function/node just implements a feature.
So I feel like we do miss a “node tool” that is just connected to any given node in the workflow (with input cardinality of 1 + no link with another trigger in the workflow, like a subworkflow would do), and leads to simpler setups overall.
It would allow to turn any node into a proper tool, even if said tools isn’t explicitly referenced here, and alleviate the need to maintain a set of many pre-wrapped tools in addition to nodes.