Python: [Bug]: FoundryAgent with tools= causes 400 error — "Not allowed when agent is specified"

[georgeollis]

Summary

The Python samples for FoundryAgent with local function tools are incorrect and produce a 400 invalid_payload error at runtime. Additionally, the intended workflow for creating a PromptAgent with tools and then invoking it via FoundryAgent is unclear.

---

Environment

• Package: agent-framework-foundry
• Python: 3.12
• Azure AI Projects SDK: latest
• Auth: AzureCliCredential

---

Reproduction

Step 1 — Agent creation via AIProjectClient (works)

from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import PromptAgentDefinition, Tool, FunctionTool
from azure.identity import AzureCliCredential

func_tool = FunctionTool(
    name="get_weather",
    parameters={
        "type": "object",
        "properties": {
            "location": {
                "type": "string",
                "description": "The city to get weather for.",
            },
        },
        "required": ["location"],
        "additionalProperties": False,
    },
    description="Get the current weather for a location.",
    strict=True,
)

tools: list[Tool] = [func_tool]

project = AIProjectClient(
    endpoint=os.getenv("FOUNDRY_PROJECT_ENDPOINT"),
    credential=AzureCliCredential(),
)

agent = project.agents.create_version(
    agent_name="TheWeatherAgent1",
    definition=PromptAgentDefinition(
        model="gpt-5.4-mini",
        instructions="Expert weather agent. Answer questions accurately and concisely.",
        tools=tools,
    ),
)

Step 2 — Invoke via FoundryAgent using tools= (fails)

Following the documented sample:

import asyncio
import os
from typing import Annotated
from agent_framework.foundry import FoundryAgent
from azure.identity import AzureCliCredential
from dotenv import load_dotenv

load_dotenv()

def get_weather(
    location: Annotated[str, "The city to get weather for."],
) -> str:
    """Get the current weather for a location."""
    return f"The weather in {location} is sunny, 22°C."

async def main() -> None:
    agent = FoundryAgent(
        project_endpoint=os.getenv("FOUNDRY_PROJECT_ENDPOINT"),
        agent_name="TheWeatherAgent1",
        credential=AzureCliCredential(),
        tools=get_weather,   # <-- as per the docs/samples
    )

    result = await agent.run("What's the weather in Paris?")
    print(f"Agent: {result}")

asyncio.run(main())

Step 2a — Try another way of using the tool decorator for FunctionTool (Not FunctonTool, part of AI Projects SDK) (fails)

import asyncio
import os
from typing import Annotated

from agent_framework.foundry import FoundryAgent
from agent_framework import tool
from azure.identity import AzureCliCredential
from dotenv import load_dotenv


@tool
def get_weather(
    location: Annotated[str, "The city to get weather for."],
) -> str:
    """Get the current weather for a location."""
    return f"The weather in {location} is sunny, 22°C."


async def main() -> None:

    agent = FoundryAgent(
        project_endpoint=os.getenv("FOUNDRY_PROJECT_ENDPOINT"),
        agent_name="TheWeatherAgent2",
        credential=AzureCliCredential(),
        tools=[get_weather]

    )
    
    result = await agent.run("What's the weather in Paris?")


if __name__ == "__main__":
    asyncio.run(main())

Error

agent_framework.exceptions.ChatClientException: 
<class 'agent_framework_foundry._agent._FoundryAgentChatClient'> 
service failed to complete the prompt: Error code: 400 - {
  'error': {
    'code': 'invalid_payload',
    'message': 'Not allowed when agent is specified.',
    'param': 'tools',
    'type': 'invalid_request_error',
    'details': [],
    'additionalInfo': {'request_id': '4b1c3bcff918421c62bc17e266ea0a06'}
  }
}

Agent in Foundry

metadata:
  microsoft.voice-live.enabled: "false"
object: agent.version
id: TheWeatherAgent2:1
name: TheWeatherAgent2
version: "1"
description: ""
created_at: 1775651912
definition:
  kind: prompt
  model: gpt-5.4-mini
  instructions: Expert weather agent demo. Answer questions about the weather accurately and concisely.
  tools:
    - type: function
      name: get_weather
      description: Get the current weather for a location.
      parameters:
        type: object
        properties:
          location:
            type: string
            description: The city to get weather for.
        required:
          - location
        additionalProperties: false
      strict: true
status: active

---

Questions

1. How should we be configuring this? It doesn't feel right that tools can't be defined alongside the agent invocation, but I appreciate that FoundryAgent expects a prompt-based agent to already exist in the service. Are there any plans for MAF to handle the creation of prompt-based agents in Foundry directly?

2. How should tool calls be defined and wired up correctly? The samples provided are incorrect and produce a 400 error — what is the intended pattern?

---

Expected Behaviour

The documented sample using tools=get_weather on FoundryAgent should either:

• Work without error (framework strips tool schemas from the request when using an agent reference), or raise a clear, actionable error message explaining

---

Package Versions

agent-framework-foundry 1.0.0

Python Version

Python 3.12

Additional Context

No response