Analyze Costs

職業

分類

除錯

Then get detailed data for a specific perspective:

Call MCP tool: harness_get
Parameters:
  resource_type: "cost_perspective"
  resource_id: "<perspective_id>"

Step 4: Get Cost Trends

Call MCP tool: harness_get
Parameters:
  resource_type: "cost_timeseries"

Step 5: Get Optimization Recommendations

Call MCP tool: harness_list
Parameters:
  resource_type: "cost_recommendation"

For detailed recommendation info:

Call MCP tool: harness_get
Parameters:
  resource_type: "cost_recommendation_detail"
  resource_id: "<recommendation_id>"

Summary statistics:

Call MCP tool: harness_get
Parameters:
  resource_type: "cost_recommendation_stats"

Step 6: Check for Anomalies

Call MCP tool: harness_list
Parameters:
  resource_type: "cost_anomaly"

To dismiss a known anomaly:

Call MCP tool: harness_execute
Parameters:
  resource_type: "cost_anomaly"
  action: "ignore"
  resource_id: "<anomaly_id>"

Step 7: Commitment Analysis

Call MCP tool: harness_get
Parameters:
  resource_type: "cost_commitment_analysis"

Also: cost_commitment_coverage, cost_commitment_utilisation, cost_commitment_savings

Report Format

## Cloud Cost Analysis

**Period:** Last 30 Days

### Summary
| Provider | Spend | vs Previous |
|----------|-------|-------------|
| AWS      | $X    | +X%         |
| GCP      | $X    | +X%         |
| Azure    | $X    | -X%         |

### Top Recommendations
1. **Rightsize <resource>** - Save $X/month (95% confidence)
2. **Convert to Reserved** - Save $X/month
3. **Delete unused resources** - Save $X/month

### Anomalies
- <date>: <service> spiked $X above expected

### Actions
- [ ] Rightsize instance X
- [ ] Purchase reserved capacity
- [ ] Clean up unused volumes

Cost Resource Types Reference

Examples

• "How much are we spending on cloud?" - Get cost_overview
• "Find $5,000 in monthly savings" - List cost_recommendations, prioritize by savings
• "Why did our bill spike last week?" - List cost_anomaly
• "Break down costs by team" - Get cost_breakdown or cost_perspective
• "Are we using our reserved instances?" - Get cost_commitment_utilisation

Performance Notes

• Gather data from all relevant perspectives before drawing cost conclusions. Partial data leads to incorrect recommendations.
• Cross-reference recommendations with anomaly data to distinguish trends from spikes.
• Quality of cost analysis is more important than speed. Verify savings estimates before presenting.

Troubleshooting

No Cost Data

• Check cloud connectors are configured and syncing
• Initial sync takes 24-48 hours
• Verify billing access permissions (Cost Explorer, Billing API)

No Recommendations

• Resources need sufficient usage history for analysis
• Small savings may be filtered out
• Check minimum running time requirements