MCP-enabled agents fail as second step in sequential YAML workflows with "Object/Array type mismatch" error

[itsdruvaraj]

Environment

• azure-ai-projects SDK version: >=2.0.0b1
• Python version: 3.x
• Date observed: January 2026

Description

When invoking an agent with MCP tools as the second step in a declarative YAML workflow, the workflow fails with an error indicating a type mismatch between Object and Array.

Error Message

Unhandled workflow failure - #validateuser_agent (InvokeAzureAgent) -> 
The requested operation requires an element of type 'Object', but the target element has type 'Array'.

Steps to Reproduce

1. Create a prompt agent without MCP tools (Agent A)
2. Create a prompt agent with MCP tools (Agent B)
3. Create a sequential YAML workflow that invokes Agent A, then Agent B
4. Run the workflow

Expected Behavior

Both agents execute successfully in sequence.

Actual Behavior

• Agent A completes successfully
• Agent B fails immediately with the type mismatch error
• The error occurs regardless of input format (text, messages, static text)

Workaround Verified

Using the same agent (without MCP) twice in sequence works correctly, confirming the workflow syntax is valid.

Minimal Reproduction Code

from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import WorkflowAgentDefinition
from azure.identity import DefaultAzureCredential

workflow_yaml = """
kind: workflow
trigger:
  kind: OnConversationStart
  id: sequential_workflow
  actions:
    - kind: CreateConversation
      id: create_conversation
      conversationId: Local.ConversationId

    - kind: InvokeAzureAgent
      id: agent_no_mcp
      conversationId: "=Local.ConversationId"
      agent:
        name: AgentWithoutMCP
      input:
        text: "=System.LastMessageText"

    - kind: InvokeAzureAgent
      id: agent_with_mcp
      conversationId: "=Local.ConversationId"
      agent:
        name: AgentWithMCPTools  # <-- This fails
      input:
        text: "Process the above."

    - kind: EndConversation
      id: end
"""

# Create and run workflow
with DefaultAzureCredential() as credential:
    with AIProjectClient(endpoint=endpoint, credential=credential) as client:
        workflow = client.agents.create_version(
            agent_name="test-workflow",
            definition=WorkflowAgentDefinition(workflow=workflow_yaml),
        )
        # Run workflow - second agent fails

Additional Context

• The MCP agent works correctly when invoked standalone
• The MCP agent works correctly when invoked as the first (and only) agent in a workflow
• The issue only occurs when MCP agent is invoked after another agent in a sequential workflow

[github-actions]

Thanks for the feedback! We are routing this to the appropriate team for follow-up. cc @dargilco @glharper @howieleung @kingernupur @nick863 @trangevi @trrwilson.

[itsdruvaraj]

This issue is seen when the agent in workflow uses MCP tool call with the allowed_tools option.