Guys i’m trying to insert input schema , but in this latest version , this feature doesn’t exist… someone know about it ? i try to find something on github but didn’t find nothing…
It looks like your topic is missing some important information. Could you provide the following if applicable.
- n8n version:
- Database (default: SQLite):
- n8n EXECUTIONS_PROCESS setting (default: own, main):
- Running n8n via (Docker, npm, n8n cloud, desktop app):
- Operating system:
This is a bug
It will likely be fixed in the next release if I were to guess
In the meantime, you can revert to an earlier version and it should fix it, use 1.73.1 in the meantime
2 Likes
I am interested in this same thing. I can’t find how to solve the issue of the input scheme that was eliminated in the new version? Can anyone provide help?
look at my reply directly above yours
it’s a bug, go to an older version if you need it now
Liam they apparently closed the problem, I really don’t understand why they changed this, now they said that all the information can be passed to the sub-node to “make it easier”, but for example we had the “extra workflow inputs” fields that were removed, in them I generally passed external information that was received via webhook, in an assertive way, so now I will be forced to pass this information to the agent for him to pass on, it doesn’t make sense, it only increases unnecessary processing of information in the agent and increases the possibilities of causing problems when it passes the parameters… I really can’t understand the rationality in taking away functionalities that give freedom to the user…
1 Like
not fixed this problem?
I’m looking for a solution to the problem, and I see the image I made and posted in the community circulating in other forums hahahahahaha. Yeah, the problem seems serious indeed. I don’t know if I should ‘downgrade’ my N8N or adapt to the new scenario. The tricky part is that every tutorial includes this extra fields thing. I guess we’ll have to adapt to the new scenario after all.
Hey all,
let me try to help with some of the confusion: We changed how the sub-workflow tool works in n8n a bit: Instead of defining a schema in the tool call, you can now define the data you expect directly in the sub-workflow: Think of it like a function in code where you define arguments.
This allows you to have a tighter fit between agent and the workflow tool and a better overview over what you need to input here.
Generally, you want to define the expected inputs in the sub-workflow trigger, like below, with whatever data the Agent (or you) should provide
When calling the tool, you can then decide if the values should be provided by the Agent or if you want to pass “hard-data” into the tool
Please let me know if think that does not make sense for any of your use cases.
3 Likes
Thank you for your explanation. I encountered the following issue. Could you help me? Thank you.
Issues encountered with the Call n8n Workflow Tool sub-node - Questions - n8n Community
But, as I can see, sub-workflow trigger node is not supporting JSON-schema format.
So, for now, how do you suggest to pass arrays of objects?
I mean, if we are using it with AI models, we need to have strict compliance between schema and input data. But with new tool node we can’t provide it.
I can’t get necessary functionallity with new tool nodes. Maybe someone has solution for this problem?
I think for now you can use the JSON example version. We will under the hood take the schema of that. But I agree that we need to bring the schema definition back.
1 Like
JSON example has a couple of problems:
- We can’t control whether properties are required.
- Values of fields in JSON properties are not descriptions, and AI model doesn’t catch it properly.
I don’t hate v2 format of workflows node (and use it for tools with simpler format), but it will be good to have JSON-schema format in the child nodes’ input.
1 Like
Hey @pashakbit, you can choose if fields are required or not when calling the tool. Also you can add a description there as well.
so with this, you cannot @Niklas_Hatje define arrays within arrays. or objects within arrays?
You can define arrays, but currently we don’t type array inside. To say it in code, we currently only support Array<any> and Record<PropertyKey, any>. I assume that is not enough for you here?
yeah, it basically means that it is limited to defining the type arrays and not a complex JSON format right? such as defining items for an array and properties for that. That was possible through the older version since it allowed for a full schema to be typed. the JSOn example schema could be used here I guess but I think it could be less reliable than just writing the schema like the older version. Also, how does the AI agent receive this as a tools information?
Yeah, we intend to bring the JSON schema back to the workflow trigger to allow that in the future. Can you specify what you mean with “how the tool receives this”?
1 Like
@Niklas_Hatje please bring back this feature. I have a use case whereby a mcp server trigger calls a n8n workflow tool. With no specify input schema option, the mcp server only accepts ‘input’ as the tool schema. So instead of for e.g. being able to pass
{
“session_id” : “example”,
“user_prompt” : “example”,
},
mcp clients would have to pass
{
“input”: “{\n "session_id": "example",\n "user_prompt": "example"\n}”
}
which is added inconvenience. I’ll have to accept all data in the subworkflow, do some extra parsing to extract the fields and so on. Please bring back the input schema. The current sub workflow ‘Workflow Input Schema’ doesn’t enforce tool schema for n8n workflow tool. 