Azure AI Deployment Manager MCP

Azure AI模型部署管理器-MCP服务器

模型上下文协议（MCP）服务器，用于管理Azure AI模型部署、监控端点和聚合Azure AI Foundry资源中的功能。

概述

Azure AI模型部署管理器为以下方面提供了一个全面的解决方案：

• 部署模型：使用可配置的SKU和实例计数将AI模型部署到Azure AI Foundry
• 列表部署：通过按资源组、项目或状态进行筛选来查询所有模型部署
• 监控端点：实时检查端点运行状况和性能指标
• 聚合能力：跨多个Azure AI资源聚合和监控AI功能
• 安全删除：删除具有内置审计跟踪和安全检查的部署

特性

1.模型部署(deploy_model)

• 将AI模型部署到Azure AI Foundry
• 配置计算SKU和实例计数
• 自动生成端点
• 实时部署跟踪

2.部署列表(list_deployments)

• 列出跨资源组的所有部署
• 按资源组、项目或状态筛选
• 全面的部署细节，包括端点
• 状态跟踪（运行、停止、失败）

3.端点监控(check_endpoint)

• 实时端点健康状态
• 性能指标（响应时间、延迟百分比）
• 可用性跟踪
• 请求速率监控

4.能力监控(monitor_capabilities)

• 汇总各地区的可用模型
• 监控计算资源（CPU、GPU、内存）
• 跟踪区域可用性
• 确定部署机会

5.安全部署删除(delete_deployment)

• 安全部署删除
• 审计跟踪日志记录
• 原因跟踪
• 防止意外删除

项目结构

azure-ai-deployment-manager/
├── src/
│   ├── __init__.py
│   ├── server.py                 # MCP server implementation
│   ├── tools.py                  # Core deployment management tools
│   ├── config.py                 # Configuration management
│   └── logging_config.py         # Logging setup
├── tests/
│   ├── __init__.py
│   ├── test_tools.py             # Unit tests for tools
│   └── test_server.py            # Unit tests for server
├── logs/                         # Application logs
├── requirements.txt              # Python dependencies
├── host.json                     # Azure Functions configuration
├── local.settings.json           # Local environment configuration
├── .env.example                  # Example environment variables
├── .gitignore
└── README.md                     # This file

先决条件

系统要求

• Python 3.11或更高版本
• pip包管理器
• Git（用于版本控制）

Azure要求

• Azure 订阅
• Azure AI Foundry项目
• 适当的Azure RBAC权限：

- Microsoft.MachineLearning/workspaces/read - Microsoft.MachineLearning/workspaces/write - Microsoft.Authorization/roleAssignments/read

环境设置

确保配置了以下环境变量：

AZURE_SUBSCRIPTION_ID=your-subscription-id
AZURE_RESOURCE_GROUP=your-resource-group
AZURE_PROJECT_NAME=your-ai-project
AZURE_TENANT_ID=your-tenant-id
AZURE_CLIENT_ID=your-service-principal-id
AZURE_CLIENT_SECRET=your-service-principal-secret
AZURE_AI_FOUNDRY_ENDPOINT=https://your-region.inference.ml.azure.com
AZURE_OPENAI_API_KEY=your-api-key

安装

1.克隆存储库

git clone https://github.com/your-org/azure-ai-deployment-manager.git
cd azure-ai-deployment-manager

2.创建虚拟环境

python -m venv venv

# On Windows
venv\Scripts\activate

# On macOS/Linux
source venv/bin/activate

3.安装依赖项

pip install -r requirements.txt

4.配置环境

复制并更新配置文件：

cp local.settings.json.example local.settings.json
# Edit local.settings.json with your Azure credentials

或者创建一个 .env 文件：

cp .env.example .env
# Edit .env with your configuration

本地测试

运行单元测试

# Run all tests
pytest

# Run with coverage
pytest --cov=src tests/

# Run specific test file
pytest tests/test_tools.py -v

# Run specific test
pytest tests/test_tools.py::TestDeploymentManager::test_deploy_model_success -v

测试输出示例

tests/test_tools.py::TestDeploymentManager::test_deploy_model_success PASSED
tests/test_tools.py::TestDeploymentManager::test_list_deployments_all PASSED
tests/test_tools.py::TestDeploymentManager::test_check_endpoint_health PASSED
tests/test_tools.py::TestDeploymentManager::test_delete_deployment PASSED
tests/test_tools.py::TestCapabilityAggregator::test_aggregate_capabilities PASSED

====== 12 passed in 2.34s ======

使用Python进行手动测试

import asyncio
from src.server import MCPServer

async def test_deployment():
    server = MCPServer()
    
    # Deploy a model
    result = await server.handle_tool_call(
        "deploy_model",
        {
            "subscription_id": "test-sub",
            "resource_group": "test-rg",
            "project_name": "test-proj",
            "model_name": "gpt-4",
            "deployment_name": "test-deployment",
            "model_version": "2024-01-01",
            "sku_name": "Standard_DS2_v2",
            "instance_count": 2,
        }
    )
    print(result)

asyncio.run(test_deployment())

使用cURL进行测试

# Start the MCP server
python -m src.server

# In another terminal, send requests
curl -X POST http://localhost:8000/tool_call \
  -H "Content-Type: application/json" \
  -d '{
    "tool": "deploy_model",
    "input": {
      "subscription_id": "test-sub",
      "resource_group": "test-rg",
      "project_name": "test-proj",
      "model_name": "gpt-4",
      "deployment_name": "test-deployment",
      "model_version": "2024-01-01",
      "sku_name": "Standard_DS2_v2"
    }
  }'

用法示例

部署模型

from src.server import MCPServer
import asyncio

async def deploy_example():
    server = MCPServer()
    
    result = await server.handle_tool_call(
        "deploy_model",
        {
            "subscription_id": "abc123def456",
            "resource_group": "my-ai-rg",
            "project_name": "ml-project",
            "model_name": "gpt-4",
            "deployment_name": "gpt4-prod",
            "model_version": "2024-01-15",
            "sku_name": "Standard_DS2_v2",
            "instance_count": 3,
        }
    )
    
    print(f"Deployment Status: {result['deployment']['status']}")
    print(f"Endpoint: {result['deployment']['endpoint']}")

asyncio.run(deploy_example())

列出所有部署

async def list_example():
    server = MCPServer()
    
    result = await server.handle_tool_call(
        "list_deployments",
        {
            "subscription_id": "abc123def456",
            "status_filter": "Running",
        }
    )
    
    print(f"Found {result['count']} running deployments")
    for dep in result['deployments']:
        print(f"  - {dep['name']}: {dep['status']}")

asyncio.run(list_example())

检查端点运行状况

async def health_check_example():
    server = MCPServer()
    
    result = await server.handle_tool_call(
        "check_endpoint",
        {
            "subscription_id": "abc123def456",
            "resource_group": "my-ai-rg",
            "endpoint_name": "gpt4-prod",
            "include_metrics": True,
        }
    )
    
    print(f"Endpoint: {result['endpoint']}")
    print(f"Status: {result['health_status']}")
    print(f"Response Time: {result['details']['response_time_ms']}ms")

asyncio.run(health_check_example())

监控功能

async def capabilities_example():
    server = MCPServer()
    
    result = await server.handle_tool_call(
        "monitor_capabilities",
        {
            "subscription_id": "abc123def456",
            "resource_groups": ["rg1", "rg2", "rg3"],
            "capability_filter": "chat",
        }
    )
    
    caps = result['aggregated_capabilities']
    print(f"Available Models: {len(caps['available_models'])}")
    print(f"Supported Regions: {len(caps['regional_availability'])}")

asyncio.run(capabilities_example())

删除部署

async def delete_example():
    server = MCPServer()
    
    result = await server.handle_tool_call(
        "delete_deployment",
        {
            "subscription_id": "abc123def456",
            "resource_group": "my-ai-rg",
            "project_name": "ml-project",
            "deployment_name": "gpt4-prod",
            "reason": "Cost optimization - consolidating to fewer deployments",
        }
    )
    
    print(f"Deployment deleted")
    print(f"Audit ID: {result['audit_trail']['audit_id']}")
    print(f"Deleted at: {result['audit_trail']['deleted_at']}")

asyncio.run(delete_example())

Azure部署

先决条件

• 已安装并配置Azure CLI
• 已安装Azure功能核心工具
• 适当的Azure权限

部署步骤

1. 创建功能应用程序

# Create resource group
az group create \
  --name azure-ai-deployment-rg \
  --location eastus

# Create storage account
az storage account create \
  --name storageaccountname \
  --resource-group azure-ai-deployment-rg \
  --location eastus

# Create Function App
az functionapp create \
  --resource-group azure-ai-deployment-rg \
  --consumption-plan-location eastus \
  --runtime python \
  --runtime-version 3.11 \
  --functions-version 4 \
  --name azure-ai-deployment-func \
  --storage-account storageaccountname

1. 部署代码

# Install Azure Functions Core Tools
npm install -g azure-functions-core-tools@4 --unsafe-perm true

# Deploy
func azure functionapp publish azure-ai-deployment-func --build remote

1. 配置应用程序设置

az functionapp config appsettings set \
  --name azure-ai-deployment-func \
  --resource-group azure-ai-deployment-rg \
  --settings \
    AZURE_SUBSCRIPTION_ID="your-subscription-id" \
    AZURE_RESOURCE_GROUP="your-resource-group" \
    AZURE_PROJECT_NAME="your-project" \
    AZURE_TENANT_ID="your-tenant-id" \
    AZURE_CLIENT_ID="your-client-id" \
    AZURE_CLIENT_SECRET="your-client-secret"

安全注意事项

身份验证和授权

1. 服务主体：使用Azure服务主体进行自动化部署
2. 管理身份：在Azure上运行时利用Azure托管身份
3. 基于角色的访问控制：实现最小权限访问控制

# Create service principal
az ad sp create-for-rbac --name azure-ai-deployment-manager

秘密管理

1. Azure密钥库：将敏感凭据存储在密钥库中
2. 环境变量：使用托管身份而不是存储机密
3. 旋转：实施定期凭证轮换

# Store secret in Key Vault
az keyvault secret set \
  --vault-name my-keyvault \
  --name api-key \
  --value your-secret-value

网络安全

1. 虚拟网络：在ViewModel中部署以实现网络隔离
2. 防火墙规则：配置防火墙限制
3. 专用端点：为Azure服务使用私有终结点

审计与合规

1. 审计日志：已登录的所有操作 logs/audit.log
2. 保留：维护合规期日志
3. 监控：设置Azure Monitor警报

审核日志条目示例：

2024-01-20 10:30:45,123 - AUDIT - DEPLOYMENT_DELETED: id=abc-123, name=old-deployment, reason=Cost optimization, deleted_by=user@example.com

日志记录

日志级别

该应用程序支持标准日志记录级别：

• DEBUG：详细的诊断信息
• INFO：一般信息性消息
• WARNING：警告信息
• ERROR：错误消息

日志文件

• logs/azure_ai_deployment.log：主应用程序日志
• logs/errors.log：仅错误
• logs/audit.log：重要业务的审计跟踪

更改日志级别

更新 LOG_LEVEL 在环境配置中：

# In local.settings.json
{
  "Values": {
    "LOG_LEVEL": "DEBUG"
  }
}

故障排除

常见问题

1.身份验证失败

Error: AADSTS700016: Application with identifier 'xxx' was not found in the directory

解决方案：验证Azure凭据和服务主体配置

az ad sp show --id your-client-id

2.超出配额

Error: Quota exceeded for SKU Standard_DS2_v2

解决方案：请求增加配额或使用不同的SKU

# Check current quotas
az vm list-skus --location eastus --output table

3.网络连接

Error: Connection timeout to Azure AI Foundry endpoint

解决方案：检查网络连接和防火墙规则

curl -I https://your-region.inference.ml.azure.com

调试模式

启用调试日志以进行故障排除：

import logging
logging.basicConfig(level=logging.DEBUG)

贡献

欢迎投稿！请遵循以下指南：

1. 创建要素分支
2. 添加新功能的测试
3. 确保所有测试通过
4. 提交拉取请求

许可证

此项目根据MIT许可证获得许可-有关详细信息，请参阅许可证文件。

支持

对于问题、疑问或贡献：

• GitHub问题：https://github.com/your-org/azure-ai-deployment-manager/issues
• 文档：https://docs.microsoft.com/en-us/azure/ai-services/
• Azure支持：https://azure.microsoft.com/en-us/support/

更新日志

版本1.0.0（2024-01-20）

• 初始版本
• 实施了5个核心工具
• 增加了全面的测试
• 创建文档
• 设置GitHub存储库

相关资源

• Azure AI Foundry文档
• 模型上下文协议规范
• Azure Python SDK
• Azure CLI参考