Azure Connector Namespace: Managed MCP servers for SRE Agent
Microsoft's Azure Connector Namespace, now in preview, hosts MCP servers that connect Azure SRE Agent to additional operational systems. While the service handles hosting and maintenance, setup still requires manual identity and connector registration, and custom server images are not yet supported.

Listen to this dispatch
Narrated by an AI-generated voice.
Microsoft has added managed MCP server hosting to its Azure ecosystem. The company's Azure Connector Namespace, now in preview, hosts Model Context Protocol (MCP) servers that connect to Azure SRE Agent, the AI service Microsoft markets for reducing operational toil — investigating incidents, automating health checks, answering change-related questions, and proposing remediations.
SRE Agent's usefulness depends on how many systems it can reach. A typical incident draws on Azure Monitor alerts, source-control history, telemetry from third-party observability platforms, and records in service-management tools. MCP servers provide the bridge: each one exposes tools that let the agent query a specific system. Microsoft ships native connections for some servers — GitHub, Datadog, New Relic, and Splunk — but for everything else, teams were left to operate their own MCP endpoints.
Connector Namespace targets that problem. Self-hosting an MCP server means deploying it, standing up HTTPS infrastructure, configuring authentication, managing downstream credentials, scaling the runtime, monitoring health, and recovering failed instances. Microsoft's argument is that none of that work adds value; the value is in the server's tools. Connector Namespace takes over the hosting and maintenance, offering a catalog that includes Azure SQL, Azure Cosmos DB, GitLab, Jira, and PagerDuty servers.
The walkthrough Microsoft published shows the setup is not zero-config. Deploying a sample SQL MCP server uses the Azure Developer CLI and requires choosing a region from a short list — West Central US, Central US, East Asia, and North Europe — plus configuring a user-assigned identity. Then the server must be registered in Azure SRE Agent's connector UI as a streamable-HTTP MCP server with managed identity authentication. Connector Namespace does not create that SRE Agent connection itself; it's a separate manual step. Teams must also grant the connector's managed identity access to the MCP server by copying tenant and principal IDs from the Azure portal and creating an access policy in the Connector Namespace portal. Only after testing the connection and selecting which server tools the agent may use does the agent gain access.
One feature is absent from the preview: bring-your-own server support, which would let teams supply a custom server image while the namespace handles hosting. Microsoft says it is "in development," with a blog post promised, but the current catalog is limited to the servers Microsoft provides.
The preview status also carries the usual caveats around supported regions and operational considerations, which Microsoft points teams to in the documentation rather than enumerating. The practical takeaway is modest: connecting SRE Agent to more operational systems is now a managed option, but it still involves identity plumbing, manual connector registration, and a catalog that doesn't yet accept custom servers. Teams evaluating the preview should budget time for the manual access-policy setup, not just the MCP endpoint.
Read the original at techcommunity.microsoft.com →