Ir para o conteúdo

Lab 030: Microsoft Foundry Agent Service + MCP

Nível: L300 Trilha: Foundry + MCP Tempo: ~90 min 💰 Custo: Assinatura Azure — custos de token gpt-4o-mini

Assinatura Azure necessária

Este lab requer uma assinatura Azure. → Guia de pré-requisitos

O que Você Vai Aprender

  • Implantar o Azure AI Foundry Hub + Projeto com um clique
  • Implantar um endpoint de modelo gpt-4o-mini
  • Criar um Agente no Foundry Agent Service
  • Conectar seu servidor MCP como uma ferramenta para o agente
  • Executar uma conversa completa do agente com chamadas de ferramentas

Implantar Infraestrutura

Opção A — Implantar no Azure (um clique)

Deploy to Azure

Isto implanta: - Azure AI Foundry Hub - AI Project (OutdoorGear Agent Project) - Storage Account + Key Vault (necessários pelo Foundry)

Localização recomendada

Use East US ou Sweden Central para melhor disponibilidade de modelos.

Opção B — Azure CLI (Bicep)

git clone https://github.com/lcarli/AI-LearningHub.git && cd AI-LearningHub

az group create --name rg-ai-labs-foundry --location eastus

# Get your user Object ID
USER_OID=$(az ad signed-in-user show --query id -o tsv)

az deployment group create \
  --resource-group rg-ai-labs-foundry \
  --template-file infra/lab-030-foundry/main.bicep \
  --parameters location=eastus userObjectId=$USER_OID

Exercício do Lab

Passo 1: Implantar gpt-4o-mini no Foundry

  1. Acesse ai.azure.com → seu projeto
  2. Model catalog → search gpt-4o-mini → Deploy
  3. Name: gpt-4o-mini — keep defaults
  4. Copie a URL do endpoint e a chave de API da página de implantação

Passo 2: Iniciar seu servidor MCP

Use o servidor do Lab 020 ou Lab 028. Para testes locais:

python mcp_product_server.py  # starts on stdio or SSE

Passo 3: Criar um agente com o SDK do Foundry

pip install azure-ai-projects azure-identity

import os
from azure.ai.projects import AIProjectClient
from azure.identity import DefaultAzureCredential

# Get these from your Foundry project settings
PROJECT_CONNECTION_STRING = os.environ["AZURE_AI_PROJECT_CONNECTION_STRING"]

client = AIProjectClient.from_connection_string(
    credential=DefaultAzureCredential(),
    conn_str=PROJECT_CONNECTION_STRING,
)

# Create the agent
agent = client.agents.create_agent(
    model="gpt-4o-mini",
    name="OutdoorGear Assistant",
    instructions=(
        "You are a helpful customer service agent for OutdoorGear Inc. "
        "Help customers find products, answer policy questions, and assist with orders. "
        "Be friendly, concise, and accurate. "
        "When you don't know something, say so honestly."
    ),
)
print(f"Created agent: {agent.id}")

Passo 4: Executar uma conversa

# Create a thread (conversation session)
thread = client.agents.create_thread()

# Add a user message
client.agents.create_message(
    thread_id=thread.id,
    role="user",
    content="Hi! I'm looking for a tent for a winter camping trip to Mt. Rainier. What do you recommend?"
)

# Run the agent
run = client.agents.create_and_process_run(
    thread_id=thread.id,
    assistant_id=agent.id,
)

print(f"Run status: {run.status}")

# Get the response
messages = client.agents.list_messages(thread_id=thread.id)
for msg in messages.data:
    if msg.role == "assistant":
        for content in msg.content:
            if content.type == "text":
                print(f"\n🤖 {content.text.value}")

Passo 5: Adicionar ferramentas MCP ao agente

# For Cloud MCP server (deployed via Lab 028):
agent_with_tools = client.agents.create_agent(
    model="gpt-4o-mini",
    name="OutdoorGear Assistant + Search",
    instructions="You are a helpful outdoor gear assistant. Use the search tool to find products.",
    tools=[
        {
            "type": "mcp",
            "server_label": "outdoorgear",
            "server_url": "https://YOUR-MCP-APP.azurecontainerapps.io/sse",
            "allowed_tools": ["search_products", "get_product", "list_categories"],
        }
    ]
)

# Now the agent can call your MCP server tools automatically
thread2 = client.agents.create_thread()
client.agents.create_message(
    thread_id=thread2.id,
    role="user",
    content="What climbing harnesses are currently in stock?"
)
run2 = client.agents.create_and_process_run(
    thread_id=thread2.id,
    assistant_id=agent_with_tools.id,
)

# Get response (agent will have called MCP tools)
messages2 = client.agents.list_messages(thread_id=thread2.id)
for msg in messages2.data:
    if msg.role == "assistant":
        print(msg.content[0].text.value)

Limpeza

az group delete --name rg-ai-labs-foundry --yes --no-wait

Próximos Passos