云优化AI代理

Cloud APIs → MCP Server → Gemini AI Agent → Streamlit Dashboard

# Install dependencies
pip install -r requirements.txt

# Copy environment template
cp .env.example .env

# Edit .env and add your Gemini API key
# GEMINI_API_KEY=your_key_here

python mcp_server.py

python gemini_agent.py

streamlit run streamlit_app.py

You are an autonomous Cloud Optimization AI Agent.

Objectives:
- Analyze cloud usage, billing, and performance metrics
- Identify inefficiencies and idle resources
- Recommend cost and performance optimizations
- Never violate SLAs
- Always estimate savings

Rules:
- Use MCP tools ONLY for data
- Never hallucinate metrics
- Output STRICT JSON
- Prefer reversible, low-risk actions

# Get cloud usage
curl -X POST http://localhost:8000/mcp/tools/get_cloud_usage \
  -H "Content-Type: application/json" \
  -d '{"provider": "gcp", "time_range": "last_30_days"}'

# Get billing data
curl -X POST http://localhost:8000/mcp/tools/get_cloud_cost \
  -H "Content-Type: application/json" \
  -d '{"provider": "gcp", "time_range": "last_30_days"}'

.
├── mcp_server.py          # MCP Server (FastAPI)
├── gemini_agent.py        # Gemini AI Agent
├── streamlit_app.py       # Streamlit Dashboard
├── requirements.txt       # Python Dependencies
├── .env.example           # Environment Template
└── README.md              # This file

from google.cloud import monitoring_v3
from google.cloud import billing_v1

# In mcp_server.py, replace mock data with:
client = monitoring_v3.MetricServiceClient()
# Fetch real metrics...

{
  "summary": "Multiple underutilized compute instances found",
  "total_potential_savings": "$185.00/month",
  "recommendations": [
    {
      "resource_id": "vm-prod-gcp-01",
      "resource_type": "compute",
      "issue": "CPU usage below 20%",
      "current_state": "CPU: 15.5% avg, Memory: 32.1%",
      "action": "Downsize machine type from n1-standard-4 to n1-standard-2",
      "estimated_savings": "$110/month",
      "risk": "low",
      "priority": "high",
      "implementation_effort": "low"
    }
  ],
  "insights": [
    "3 compute instances show underutilization",
    "Storage costs are within acceptable range"
  ]
}