Microsoft Adds Cost Management Tools to Azure ARM MCP Server

Microsoft has added Cost Management tools to its Azure Resource Management (ARM) MCP server, allowing AI agents to query cost data, price sheets, budgets, and savings recommendations within Azure workflows. A small set of cost tools — historical spend queries, AKS cost queries, retail pricing, and price sheet downloads — is enabled by default. An optional toolset covers forecasting, budgets, alerts, reservations, and Savings Plans, activated by adding an x-mcp-toolset: CostManagement header to the server configuration.

The default tools are deliberately narrow. According to the announcement from Microsoft's Azure Governance and Management team, the built-in set supports "query" and "pricing" scenarios: query_costs, query_aks_costs, get_retail_prices, start_pricesheet_download, and get_pricesheet_status. Everything else requires opt-in. The optional toolset adds list_dimensions, forecast_costs, budget creation and review, alert listings, benefit recommendation and utilization queries, and reservation transaction history.

Cloud architecture and cloud economics are intertwined, but the data for each lives in separate tools. The ARM MCP server already exposes resource management; the new cost layer lets an agent answer questions like "what were my top cost drivers this month?" or "how am I doing against my budgets?" in the same context.

The announcement frames several use cases. Before deployment, an agent could estimate costs for proposed resources, compare VM sizes against an ARM template, and factor in negotiated pricing. After deployment, it could explain spending trends. For AKS environments, it can query cost by cluster and namespace and compare active versus idle capacity. Microsoft also claims the optional tools surface Advisor cost recommendations, such as Savings Plans that would save at least a specified amount.

Those examples are illustrative prompts from the announcement, not demonstrated results. What is established is that the tools exist, what they query, and how they're configured.

Deployment has some friction. The default tools work out of the box through the ARM MCP server, but the optional toolset requires editing an mcp.json file — either per workspace or globally in VS Code — or the Copilot CLI config, and restarting the MCP server to pick up the change. Authentication requires an Azure account with appropriate permissions and a GitHub Copilot subscription. Microsoft points users to a troubleshooting guide for common sign-in issues.

Two practical observations. First, the pricing tools distinguish between retail prices and negotiated rates, returning the latter "where available" — agent-generated cost estimates are only as good as whatever price data the tenant exposes. Second, the opt-in model makes sense for MCP servers, where every tool is a potential attack surface for a model with credential access; a cost tool that can create budgets is more dangerous than one that reads spend.

Microsoft says it is actively expanding these capabilities, with "additional supported tools and skills" planned. The documentation and feedback links are live, and bug reports route to the ARM MCP GitHub repository. Where this goes next — and how much of the cost-analysis work Microsoft keeps on the server side versus pushing to agents — is still open.