Microsoft lets AI agents read Azure data after one sign-in
Each agent step uses the access of the user, and the agent cannot get more than that user can see.
Microsoft released one Azure Monitor server for the Model Context Protocol, a standard that lets AI agents use tools. Agents such as GitHub Copilot can read Azure metrics and logs, and find and change dashboards. A user must have one install connection and a sign-in. No Grafana dashboard tool and no server that the user installs is necessary.
How it works
The server has 21 tools. 20 of them only read. One tool changes a kept dashboard that the user can change. It does not keep a change if the dashboard changed after the agent read it.
The test
The author gave Copilot one task about storm damage for each person in each US state. Copilot used 20 steps in 1 minute 25 seconds, and 12 of the steps used the server. It found the data cluster, used 2 tables and gave a ranking. It also told the author that the population table has no year field.
What is not known
The author has one test on a public sample data set. There is no test on large systems. Microsoft tells that this server does not replace the server in each Azure Managed Grafana instance. That server is the right one for teams that use the same data sources.
Insight: Your sign-in sets the limit of your agent
Anthropic released Claude Dashboards in beta, and a user can move a dashboard to tools such as Grafana. Microsoft lets agents read and change dashboards in Azure, with no data source to set up. This seems to be a move to make the Microsoft sign-in system the control point for agents. Look for which agent tools use the same sign-in. Wide access roles give wide access to agents.
Sources
- Introducing Azure Monitor dashboards with Grafana MCP: Telemetry for AI Agents, Signed In as Youtechcommunity.microsoft.com
Posted