Your FinOps data, one question away.
A CloudMonitor MCP server is on the roadmap. It will let Claude, ChatGPT, Copilot, and any other MCP-compatible assistant query cost, budget, and anomaly data straight from your Fabric workspace, without anyone opening a dashboard.
A model context protocol (MCP) server turns CloudMonitor's reports into a tool an AI assistant can call directly. Instead of building a dashboard link or exporting a CSV, an engineer or FinOps lead asks a question in the assistant they already have open, and the answer comes back from your own governed data, not a model's guess.
- Will answer natural-language questions about cost, budgets, and anomalies from any MCP-compatible assistant.
- Will connect over remote MCP with OAuth, the same sign-in used for the CloudMonitor app.
- Will scope every answer to the asking user's existing role and cost-group permissions.
- Will read from the same governed Fabric workspace as the reports, so nothing is a second copy or a stale export.
Natural-language queries
Ask instead of navigating.
The MCP server will expose CloudMonitor's core reporting as callable tools, so questions like "what did the platform team spend on Azure last month" or "which anomalies are still open" get answered inline in the assistant, with the same numbers the CloudMonitor app would show.
- Cost and trend queries by team, cost group, or resource
- Anomaly and budget status without opening the inbox
- Savings and waste recommendations on request
Read-only, permission-scoped
The same access, not more.
The server will be read-only and will enforce the same role-based and cost-group access controls as the CloudMonitor app. An assistant answering on someone's behalf sees exactly what that person is allowed to see, and nothing gets written back.
- OAuth sign-in, no shared API key to manage
- Existing role and cost-group scoping, not elevated access
- Read-only: no configuration or data changes from a client
Most FinOps questions are quick ones: what did we spend, is anything anomalous, is a budget close to breaching. Today they mean opening CloudMonitor to find out. An MCP server removes that step. The assistant already open in an engineer’s editor or a FinOps lead’s browser tab becomes a client that can ask CloudMonitor directly and get a governed answer back.
This connector is in planning for Q4 2026, alongside Finn, CloudMonitor’s own FinOps data agent. If your team already leans on Claude, Copilot, or another MCP-compatible assistant, the beta team would like to know which questions you’d ask it first.
More integrations
Other integrations like this.
Be first to connect MCP Server.
Join the beta and tell the team what you need from this connector.
