Azure Pricing Agent Skill

Azure Pricing Skill

Fetch live Azure retail pricing data via the Azure MCP Server pricing tool and turn it into actionable cost estimates, region comparisons, and deployment-level forecasts.

When to Use

• "How much does Standard_D4s_v5 cost in uksouth?"
• "Estimate the cost of this Bicep/ARM/Terraform template"
• "Compare VM pricing between uksouth and westeurope"
• "Show me Reservation vs Consumption pricing for SQL Database"
• "What are the cheapest storage SKUs in uksouth?"
• "Include savings plan pricing for this compute tier"
• Any request for Azure pricing, cost estimates, or SKU comparisons

Required Tooling

• MCP tool: Azure pricing tool provided by the Azure VS Code extension (ms-azuretools.vscode-azure-github-copilot). No mcp.json configuration is required — the tool is registered automatically by the extension.

> Before invoking the pricing tool, always run tool_search_tool_regex with pattern pricing to discover the exact tool name. Do not guess or hardcode the name.

Tool Parameters

| Parameter | Required | Description |

|-----------|----------|-------------|

| sku | Optional* | ARM SKU name (e.g. Standard_D4s_v5, Standard_E64-16ds_v4) |

| service | Optional* | Azure service name (e.g. Virtual Machines, Storage, SQL Database) |

| region | Optional* | Azure region slug (e.g. eastus, westeurope, westus2) |

| service-family | Optional* | Service family (e.g. Compute, Storage, Databases, Networking) |

| price-type | Optional* | Consumption, Reservation, or DevTestConsumption |

| include-savings-plan | Optional | true to include savings plan pricing (uses preview API; mainly applies to Linux VMs) |

| filter | Optional* | Raw OData filter expression for advanced queries |

| currency | Optional | Currency code. Skill default: GBP — always pass this explicitly. The underlying tool defaults to USD if omitted. Other common values: USD, EUR |

> *At least one filter parameter is required per call.

> Gotchas:

> - SavingsPlan is NOT a valid price-type value. To get savings plan rates use include-savings-plan: true alongside a Consumption query.

> - Prefer querying with a specific SKU rather than a broad service name alone — results will be more targeted and useful. Querying by service name without a SKU is valid when the user explicitly wants a full listing of available SKUs.

> - SKU name spacing varies by service. The sku parameter is format-sensitive. For App Service, P2v3 returns no results but P2 v3 (with a space) works. If the sku parameter returns empty, fall back to an OData filter with skuName eq '...' to match exactly as the API stores it.

> - Reservation retailPrice values are lump-sum totals, not hourly rates. Despite unitOfMeasure: "1 Hour", Reservation price rows return the total commitment cost (annual or 3-year). Divide by 8,760 (1-year) or 26,280 (3-year) to get an hourly equivalent for comparison.

> - SQL Database compute and storage are separate meters. You need two calls: one for compute (e.g. 4 vCore under SQL Database Single/Elastic Pool General Purpose - Compute Gen5) and one for storage (SQL Database Single/Elastic Pool General Purpose - Storage). A single query returns both if you don't filter by productName.

> - SQL Database skuName in the API uses plain English, not ARM format. The ARM SKU GP_Gen5_4 maps to API skuName: "4 vCore" under productName containing General Purpose - Compute Gen5. Filter by both skuName and productName to avoid Business Critical or DC-Series rows returning alongside General Purpose.

> - Spot pricing requires the filter parameter, not price-type. Spot is not a price-type value — use filter: "contains(meterName, 'Spot') and armSkuName eq '<sku>' and armRegionName eq '<region>'" with price-type: Consumption. Live-tested: Standard_D4s_v5 in uksouth returns Linux Spot at £0.0213/hr and Windows Spot at £0.039/hr (vs £0.164/hr and £0.300/hr standard — ~87% saving).

> - isPrimaryMeterRegion eq true in OData filter: When using targeted sku + region queries, the MCP tool already returns only primary meter region rows. The isPrimaryMeterRegion eq true filter expression is useful when writing broad OData-only queries (without a locked region) that might otherwise return duplicate rows for the same SKU across meter variants.

> - include-savings-plan: true is incompatible with the filter parameter. When include-savings-plan: true is set alongside an OData filter, the tool returns empty results. To retrieve savings plan data, use sku + region parameters only. The savingsPlan field is a nested array ([{"term": "1 Year", "unitPrice": ...}, {"term": "3 Years", ...}]) present only on Linux compute rows — Windows VM rows never include savings plan data.

> - DevTestConsumption waives the Windows OS license cost on VMs. Windows VMs queried with price-type: DevTestConsumption return at Linux-equivalent prices (e.g. Standard_D4s_v5 uksouth: £0.164/hr, same as Linux PAYG). The Windows license is free under Dev/Test subscriptions. Always flag this when estimating costs for Dev/Test Windows VMs.

> - Reservation queries always return both 1-year and 3-year rows together. There is no reservation-term parameter — it is silently ignored. To isolate a single term, combine price-type: Reservation with filter: "reservationTerm eq '1 Year'" or filter: "reservationTerm eq '3 Years'". Parameters top, skip, and meter-name are also silently ignored by the MCP tool.

> - ⚠️ Azure OpenAI / AI services cannot be queried via this MCP tool. The Azure MCP pricing tool returns a 500 error for any query that resolves to the AI + Machine Learning service family — including service: "Foundry Models", service: "Azure OpenAI", service-family: "AI + Machine Learning", and OData filters matching those services. This is a known MCP tool limitation. For AI model pricing (GPT-4o, text-embedding, etc.), use the Azure Pricing Calculator or the Azure OpenAI pricing page directly.

Workflow

Scenario 1: Single SKU / Service Price Lookup

• Extract the SKU, service name, or region from the user's request.
• Prefer a specific SKU or tier. If the user hasn't provided one and a full SKU listing isn't what they're after, ask for clarification before calling the tool.
• Call the pricing tool with the identified parameters.
• Present the result as a formatted price table (see Output Format below).

Example prompts to tool:

• SKU in region: { "sku": "Standard_D4s_v5", "region": "uksouth", "currency": "GBP" }
• With reservation comparison: { "sku": "Standard_D4s_v5", "region": "uksouth", "price-type": "Reservation", "currency": "GBP" }
• With savings plan (Linux VMs only): { "sku": "Standard_D4s_v5", "region": "uksouth", "include-savings-plan": true, "currency": "GBP" }

> Note: Calling with service alone (e.g. "service": "Virtual Machines") without a sku returns all SKUs for that service — only do this if the user explicitly wants a full listing.

Scenario 2: Template Cost Estimation (Bicep / ARM / Terraform)

Pre-estimation: ask about usage patterns before querying — for consumption-based services, the cost varies wildly based on scale. Before calling the pricing tool, if the template contains any of the following, ask the user:

| Service | Ask about |

|---------|----------|

| Azure Functions | Expected invocations/month, average execution duration (ms), memory allocation (GB) |

| Cosmos DB (serverless) | Expected RU consumption/month |

| Container Apps | Expected request volume, scale-to-zero behaviour, vCPU/memory allocation |

| Azure OpenAI | Note: AI service pricing is not queryable via this MCP tool — direct users to the Azure OpenAI pricing page |

For always-on services (VMs, App Service, SQL Database), proceed directly with the pricing query.

• Parse the template to identify all resource types and their SKUs/tiers.
• For each resource, call the pricing tool with the appropriate sku and/or service + region.
• Aggregate results into a total monthly and annual cost estimate (monthly × 12).
• Flag any resources where pricing could not be retrieved.

Common Terraform resource mappings (examples — not exhaustive):

• azurerm_app_service_plan → sku_name (e.g. P1v3)
• azurerm_linux_virtual_machine / azurerm_windows_virtual_machine → size (e.g. Standard_D4s_v5)
• azurerm_mssql_database → sku_name (e.g. GP_Gen5_4)
• azurerm_storage_account → account_tier + account_replication_type (e.g. Standard_LRS)
• azurerm_cosmosdb_account → throughput settings
• azurerm_kubernetes_cluster → default_node_pool.vm_size × node count
• azurerm_redis_cache → sku_name + family + capacity
• azurerm_servicebus_namespace → sku
• azurerm_api_management → sku_name
• azurerm_container_app_environment + azurerm_container_app → workload_profile / cpu + memory

Common Bicep/ARM resource mappings (examples — not exhaustive):

• App Service Plans → SKU name (e.g. P1v3, P2v3)
• Virtual Machines → VM size (e.g. Standard_D4s_v5)
• Azure SQL → Service tier + compute size (e.g. GP_Gen5_4)
• Storage Accounts → SKU + redundancy (e.g. Standard_LRS, Premium_ZRS)
• Cosmos DB → Provisioned throughput (RU/s)
• Azure Kubernetes Service → Node VM size × node count
• Container Apps → workload profile + vCPU/memory allocation
• API Management → SKU tier (e.g. Developer, Premium)

> ⚠️ Azure Hybrid Benefit (AHB): The retail pricing API returns pay-as-you-go rates and never reflects Azure Hybrid Benefit discounts. AHB can reduce costs by 40%+ for Windows VMs and SQL Server workloads. Always flag this when estimating costs for Windows or SQL resources, and direct users to the Azure Hybrid Benefit calculator for accurate figures.

Scenario 3: Region Comparison

• Extract the service/SKU from the request.
• Call the pricing tool once per region to compare.
• Present a comparison table sorted by price (ascending).

Recommended region set for comparisons:

• uksouth, ukwest, westeurope, northeurope, eastus, eastus2 etc

Scenario 4: Price Type Comparison (Consumption vs Reservation)

• Fetch Consumption pricing first.
• Fetch Reservation pricing for the same SKU/service (set price-type: Reservation).
• If requested, include savings plan by passing include-savings-plan: true on the Consumption call (savings plan pricing is surfaced as a nested array on each result — applies mainly to Linux VMs).
• Present a side-by-side comparison with calculated savings percentages.

> Reminder: SavingsPlan is not a valid price-type. Always use include-savings-plan: true flag for savings plan rates.

Scenario 5: Advanced / OData Filter

Use the filter parameter for complex queries:

• Specific meter: meterId eq 'abc-123'
• Price range: retailPrice le 0.10
• Combined: serviceName eq 'Storage' and skuName eq 'LRS' and armRegionName eq 'eastus'
• Spot pricing: filter: "contains(meterName, 'Spot') and armSkuName eq 'Standard_D4s_v5' and armRegionName eq 'uksouth'" with price-type: Consumption — returns Linux and Windows Spot rows separately
• Primary meter only (broad queries): filter: "serviceName eq 'Virtual Machines' and armRegionName eq 'uksouth' and isPrimaryMeterRegion eq true" — avoids duplicate rows when not using a specific SKU

> Note: Avoid OData filters that resolve to serviceFamily eq 'AI + Machine Learning' — the MCP tool will return a 500 error. See AI services note in Gotchas above.

Output Format

> These are reference templates — adapt the format to context. A quick inline answer may need only a line or two; a full architecture document warrants the complete table format.

Single Price Lookup

## Azure Pricing: [Service/SKU]

| Field | Value |
|-------|-------|
| Service | Virtual Machines |
| SKU | Standard_D4s_v5 |
| Region | UK South |
| Price Type | Consumption |
| Retail Price | [from tool]/hour |
| Monthly Est. | ~[from tool]/month (730 hrs) |
| Annual Est. | ~[from tool]/year |
| Currency | GBP |

**Spot:** [from tool]/hour ([x]% saving vs Consumption) — interrupt-tolerant workloads only
**Savings Plan (1-year):** [from tool]/hour ([x]% saving vs Consumption)
**1-Year Reservation:** [from tool]/hour ([x]% saving)
**3-Year Reservation:** [from tool]/hour ([x]% saving)

Template Cost Estimate

## Deployment Cost Estimate
**Template**: [filename or description]
**Region**: uksouth
**Currency**: GBP

| Resource | SKU / Tier | Monthly Cost | Annual Cost |
|----------|-----------|-------------|-------------|
| App Service Plan | P2v3 | £240.00 | £2,880.00 |
| Azure SQL Database | GP_Gen5_4 | £304.00 | £3,648.00 |
| Storage Account | Standard_LRS | £15.00 | £180.00 |
| Application Insights | Pay-as-you-go | ~£8.00 | ~£96.00 |
| **Total** | | **~£567/month** | **~£6,804/year** |

> Note: Estimates based on retail (pay-as-you-go) pricing. Reserved instances or savings plans can reduce this by 20-70%.

**Cost Reduction Opportunities:**
- Switch App Service Plan to 1-year reservation: save ~£72/month
- SQL Database 1-year reservation: save ~£122/month
- **Total potential savings with reservations: ~£194/month (34%) / ~£2,328/year**

Region Comparison Table

## Region Price Comparison: Standard_D4s_v5 (Consumption)

| Region | Price/Hour | Monthly Est. | vs Cheapest |
|--------|-----------|-------------|-------------|
| UK South | £0.158 | £115 | baseline |
| UK West | £0.165 | £121 | +4% |
| North Europe | £0.166 | £121 | +5% |
| West Europe | £0.173 | £126 | +10% |
| East US | £0.158 | £115 | 0% |
| Australia East | £0.212 | £155 | +34% |

**Recommendation**: `uksouth` offers the lowest cost within the UK for this SKU.

Integration with Other Skills

| Scenario | Combine With |

|----------|-------------|

| Full architecture cost estimate | architecture-design skill |

| Identify expensive resources in existing deployments | cost-optimization skill |

| WAF cost pillar review | waf-assessment skill |

| IaC with cost-aware service selection | architecture-design + azure-pricing |

| Terraform cost estimation | Parse .tf files, call pricing tool per resource, aggregate in GBP |

Common Service Name Reference

The table below covers frequently used services — use the service name exactly as shown. For services not listed, try the exact Azure portal display name or use the filter parameter with an OData expression.

| Azure Service | service Parameter Value | Notes |

|---------------|--------------------------|-------|

| Virtual Machines | Virtual Machines | |

| App Service | Azure App Service | |

| Azure SQL Database | SQL Database | |

| Azure Cosmos DB | Azure Cosmos DB | |

| Azure Kubernetes Service | Azure Kubernetes Service | |

| Azure Functions | Azure Functions | |

| Storage (Blob/Queue/Table) | Storage | |

| Azure Cache for Redis | Redis Cache | API service name differs from portal display name |

| Service Bus | Service Bus | |

| Event Hubs | Event Hubs | |

| API Management | API Management | |

| Application Gateway | Application Gateway | |

| Azure Front Door | Azure Front Door | |

| Log Analytics | Log Analytics | |

| Application Insights | Application Insights | |

| Azure Container Apps | Azure Container Apps | ⚠️ Cannot use service parameter — tool returns 400 (no ARM SKU). Use filter: "serviceName eq 'Azure Container Apps' and armRegionName eq '<region>'" instead. Multi-unit pricing (vCPU/sec, GiB-sec, requests/1M) — always ask about usage before estimating |

| Azure OpenAI / Foundry Models | Foundry Models | ⚠️ Not queryable via MCP tool — AI + Machine Learning service family causes 500 error. Use Azure OpenAI pricing page directly |

Price Type Guide

| Price Type | When to Use |

|-----------|-------------|

| Consumption | Default pay-as-you-go; no commitment |

| Reservation | 1-year or 3-year committed use; 20-72% off |

| DevTestConsumption | Dev/Test subscriptions; discounted non-prod rates |

Error Handling

If the pricing tool returns no results:

• Consult SKU Name Quirks — the API skuName often differs from the ARM/portal/IaC name (e.g. P2v3 → P2 v3, GP_Gen5_4 → 4 vCore, C1 → C1 Cache Instance).
• Try broadening the query (remove SKU, keep only service + region).
• Verify the service name matches the Common Service Name Reference above.
• Try the filter parameter with an OData expression — this is more reliable than the sku parameter for services where the API skuName format differs from the portal/ARM name (e.g. App Service, SQL Database).
• If skuName format is unknown, query without it first to see what skuName values are returned, then narrow.
• If service name is unknown, use service-family alone (e.g. service-family: Compute, service-family: Databases) to discover what serviceName values exist in that family, then re-query with the correct name.
• Inform the user if pricing is unavailable for a specific SKU and suggest the nearest alternative.

If the pricing tool returns a 500 error:

• Check whether the query would resolve to serviceFamily eq 'AI + Machine Learning'. If so, this is a known MCP tool limitation — redirect the user to the Azure Pricing Calculator or Azure OpenAI pricing page.

Reference Documentation

• Azure Pricing MCP Tool
• Azure Retail Prices API
• Azure Pricing Calculator
• Azure Reservations
• Azure Savings Plans
• Azure OpenAI Pricing *(use directly — not queryable via MCP tool)*
• Cost Estimation Formulas *(service-specific formulas, Spot pricing reference, pre-estimation guidance)*
• SKU Name Quirks *(ARM → API skuName mappings for App Service, SQL Database, Redis, Storage, APIM, Container Apps)*