
Microsoft has made the Azure DevOps Remote MCP Server generally available, but the hosted endpoint still does not support Claude Desktop, Claude Code, ChatGPT, or Cursor. The reason is not in those tools, according to Microsoft’s explanation in the release notes and follow-up commentary, but in the authentication layer: the service depends on Microsoft Entra capabilities those clients do not yet meet.
A hosted Azure DevOps endpoint for AI assistants
The remote server gives AI assistants a direct path into Azure DevOps work items, pull requests, repositories, and pipelines without requiring developers to install or run anything locally. Microsoft exposes the service at https://mcp.dev.azure.com/{organization} over streamable HTTP, and a supported client connects by adding a single server entry to mcp.json.
In practice, that makes the setup look simple:
{
"servers": {
"ado-remote-mcp": {
"url": "https://mcp.dev.azure.com/{organization}",
"type": "http"
}
},
"inputs": []
}Azure DevOps hosts the endpoint and authenticates through Microsoft Entra, which is central to both the promise and the current limitations of the release.
Why Claude, ChatGPT, and Cursor are not supported yet
Dan Hellem, product manager for Azure Boards, Repos, and Wiki, said support depends on a client’s ability to authenticate with Entra. He said clients such as Claude Desktop, Claude Code, ChatGPT, and Cursor need support for dynamic OAuth client registration or Client ID Metadata Documents in Entra before they can connect. Microsoft is working with the Entra team to enable that capability, but there is no timetable for when that work will land.
Until then, customers using those tools are expected to keep using the local server. Microsoft says the local MCP Server remains available and that it recently consolidated the local toolset to align it with the remote server, with a commitment to keep the two in parity while the Entra work continues.
Microsoft’s first-party clients get access now
Not every AI tool is blocked. Microsoft says a number of its own products can already connect without extra onboarding, including Visual Studio Code with GitHub Copilot, Microsoft Foundry through its tools catalog, Copilot Studio, Visual Studio, the GitHub Copilot CLI, and the GitHub Copilot app.
The distinction is important for teams trying to standardize on one assistant. Microsoft’s own ecosystem can use the hosted endpoint today, while third-party agents remain tied to the local deployment model.
What Microsoft says the security model buys
Farhan Shahnewaz, an AI and cloud solutions engineer at Microsoft, argues that the Entra requirement is a feature rather than a bug. In his walkthrough, he says there is no personal access token sitting in a configuration file waiting to leak. Instead, Entra ensures the assistant inherits exactly the developer’s permissions and nothing more, which he says is the first question security teams tend to ask.
That security framing is likely to matter for enterprises weighing whether to connect an AI assistant to source control and delivery pipelines. A hosted server removes the burden of running and maintaining a local instance, but it also shifts the trust boundary to the identity provider and the supported client.
A spec change makes the timing more awkward
The GA release arrives just one week after the MCP 2026-07-28 specification. That version reordered client authentication mechanisms so that pre-registered clients come first, Client ID Metadata Documents come next, and Dynamic Client Registration is now a deprecated fallback scheduled for removal after summer 2027.
According to the report, Entra does not yet support the two mechanisms the updated specification relies on, and one of those mechanisms is already on a path to removal from the protocol. That leaves Microsoft with a standards mismatch to resolve if it wants broader support across the client ecosystem.
One limitation is permanent
There is also a restriction Microsoft does not appear to plan around. An Entra tenant must back the organization, which means standalone organizations using Microsoft accounts are not supported. For some teams, that is a hard boundary rather than a temporary compatibility issue.
So while the remote server is generally available, it is not universally available in the way the word “remote” might suggest. The supported path depends on both the organization’s identity setup and the AI client’s authentication capabilities.
What this means for DevOps teams
The practical difference is clear. Teams already using Microsoft’s first-party tools can point them at a hosted Azure DevOps MCP endpoint and avoid managing their own server. Teams standardizing on Claude Code or Cursor do not get that benefit yet and must continue hosting locally.
That matters most in day-to-day operational work such as pipeline triage. Shahnewaz describes using the assistant to inspect a two-stage pipeline behind a small ASP.NET Core app, asking which stage failed and why instead of opening multiple browser tabs and digging through thousands of log lines. The hosted model removes setup work for developers who can use it, but teams outside Microsoft’s supported client list still carry the maintenance overhead.
Key takeaways from the GA release
- The Azure DevOps Remote MCP Server is now generally available.
- It exposes Azure DevOps work items, pull requests, repositories, and pipelines through a hosted endpoint.
- Claude Desktop, Claude Code, ChatGPT, and Cursor are not supported yet because of Entra authentication requirements.
- Microsoft says it is working with the Entra team, but no timeline has been announced.
- The local MCP Server remains available and Microsoft says parity will be maintained.
The release underscores a broader reality for enterprise AI tools: a protocol can standardize how clients discover and call tools, but it cannot by itself harmonize every identity provider’s policy or every vendor’s authentication roadmap.
Source: Original report
Was this helpful?
Explore more: DevOps Services More Cloud & DevOps Tech News
Last Modified: August 23, 2026 at 1:53 am
0 views
