Agent 365 can now block single MCP tools at Azure’s gateway
Microsoft lists the Agent 365 integration with Azure API Management as generally available, giving admins one place to inventory and block MCP servers, and on one tier, individual tools.

Microsoft has marked its Agent 365 integration with Azure API Management and Azure AI Gateway as generally available. Administrators can now govern Model Context Protocol (MCP) servers from Agent 365 while the gateway enforces the rules at runtime.
The Azure Updates entry published on November 17 carried only its title when we read it, so the details below come from Microsoft’s earlier preview notice and its Learn documentation. Those pages predate the GA label, and the Learn article we read still describes the integration as a preview. Readers should check the current docs before relying on any single detail.

What admins can control from Agent 365
According to Microsoft Learn, Agent 365 sets governance intent, inventory and access decisions, while API Management enforces policy inline on traffic to MCP servers and tools. Agent 365 discovers resources of type Microsoft.ApiManagement in a tenant, then pulls in the MCP servers and tools they expose, plus metadata such as endpoints, access scopes and service identities.
Four things are listed: a central inventory of gateway-managed MCP servers, blocking and unblocking of whole servers, tool-level blocking, and tool activity data for auditing and incident response. Microsoft says Agent 365 also works with Microsoft Defender and Microsoft Purview to check tool requests and results for security and sensitive-data risks.
Which of these you get depends on the tier. Per Microsoft’s capability table, the v2 tiers of API Management offer the inventory and server-level blocking. Tool-level blocking, activity observability and threat protection need the AI Gateway tier. In the October preview notice, Microsoft said that tier is still in preview. API Management v1 tiers are not supported.
Setup needs a tenant admin and a static tool list
To connect, an organization needs an Agent 365 tenant, a supported API Management instance and MCP servers registered in it. A tenant-level administrator must give one-time consent, which grants Agent 365 the A365 MCP Gateway Policy Operator Azure role. That role lets the service read API Management resources and read and write API-scoped policies, including in future subscriptions. After that, governance actions run under Agent 365’s own app-only identity, so individual admins need no separate Azure permissions.
Tool-level control has a catch. Each MCP server needs a static list of allowed tools configured, and Microsoft says servers with dynamic tool lists cannot be governed at tool level from Agent 365. For those, the AI Gateway portal has to be used instead.
Blocked calls fail loudly, and routing is still on you
The blocking mechanics are specific. Blocking a server inserts a managed return-response policy at the start of the API’s inbound policy, and the gateway then answers with HTTP 403. Blocking a single tool, such as the delete_file example in the docs, returns a 200 response carrying a JSON-RPC error marked isError, with a message telling the agent not to retry. Microsoft says this is meant to make agents aware of the policy rather than loop on it.
The angle that matters here is safety. Agents that can call tools can delete files or send data, and until now many teams could only allow or deny an entire MCP server. Disabling one risky tool while leaving the rest usable is a finer control, and it sits in the same console that already tracks agents. Microsoft has said Agent 365 itself has been generally available for commercial customers since May 1, 2026, licensed per user, according to the consultancy innFactory, which means the licensing is separate from Azure gateway costs. We found no pricing for the gateway integration in the sources we read.
There is also a limit the docs state plainly: Agent 365 does not automatically route agents through a customer-hosted gateway. A developer must point each agent at the gateway’s MCP endpoint, and Microsoft advises locking MCP servers down so the gateway is the only permitted caller, using private networking where supported. An agent configured to reach a server directly would bypass every control described above.
What the sources do not settle
The GA status rests on Microsoft’s own update title. Because the AI Gateway tier was still labeled preview in the sources we read, it is not clear whether the tool-level features are covered by GA or only the v2 tier capabilities. Regional availability, pricing for the gateway tier and the date the documentation will be updated are not stated in anything we could read. All capability descriptions are Microsoft’s own and have not been independently tested.
The release lands in the same week as Microsoft’s Ignite announcements, which we covered in our Ignite report, and as the company previewed other agent controls such as Agent Grid and an Agent Sandbox cluster extension for AKS.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Leo MartinsAI Tools, Prompts & Practical AI ExpertLeo Martins is the AI Tools & Practical AI Expert at Aiholics, focused on helping readers use artificial intelligence to improve productivity, creativity and everyday work.



