Issue while executing ADO mcp tools

[ShubhamMukherjeeTR]

As shown in above image, I am getting error during agent execution step 1:1 validation error for dynamicmodel
Field required [type=missing, input_value={'project': 'taxprof', 'workItemType': 'Bug'}, input_type=dict]
For further information visit https://errors.pydantic.dev/2.11/v/missing.
Can anybody inform how to resolve this Pydantic error ?

[ShubhamMukherjeeTR]

This error is occuring for wit_create_work_item tool only in https://github.com/microsoft/azure-devops-mcp

[pietrozullo]

Ok it is pretty tough for me to test this since I do not have the adobe project. But from the source of the server:

server.tool(
    WORKITEM_TOOLS.create_work_item,
    "Create a new work item in a specified project and work item type.",
    {
      project: z.string().describe("The name or ID of the Azure DevOps project."),
      workItemType: z.string().describe("The type of work item to create, e.g., 'Task', 'Bug', etc."),
      fields: z.record(z.string(), z.string()).describe("A record of field names and values to set on the new work item. Each key is a field name, and each value is the corresponding value to set for that field."),
    },

the LLM is confused about what it should put in the field, which is of Record type.

In fact you see that the input from the LLM is {'project': 'taxprof', 'workItemType': 'Bug'} and fields is missing, therefore the error.
Try to specify what to put there, the model might not be smart enough to understand that it should put an empty dict if you say nothing.

[ShubhamMukherjeeTR]

Here, ADO is Azure Devops and not Adobe. I tried passing fields as a input like :

Create a bug in ado where : {
"project": "taxprof",
"workItemType": "Bug",
"fields": {
"System.Title": "AI new task",
"System.Description": "This is a description of the task."
}
}

But still same error. Is it possible to handle in mcp-use where step 1:1 validation error for dynamicmodel here even if this is failing, it will move ahead and try to execute the job ?

[pietrozullo]

Ok, can you share the tool call logs ? This is your query, but what did the LLM send to the tool ? Same as before ?

Anyway, the continue/retry logic is a very good idea, I will open an issue.

[ShubhamMukherjeeTR]

Logs are same as before