Microsoft launches the hosted Azure DevOps Remote MCP Server, enabling AI assistants to securely access Azure DevOps projects without requiring local installation or self-hosting.

Microsoft has announced the general availability (GA) of the Azure DevOps Remote MCP Server, making it easier for developers and organizations to connect AI assistants directly to Azure DevOps projects through a fully managed service. The announcement introduces a hosted endpoint that eliminates the need to install or manage a local MCP server, allowing supported AI tools to securely access project data using the Model Context Protocol (MCP) over HTTP.

The release represents Microsoft’s latest effort to integrate AI more deeply into software development workflows by enabling assistants to work with repositories, work items, pull requests, and pipelines while maintaining enterprise authentication through Microsoft Entra.

What Microsoft Announced

The Azure DevOps Remote MCP Server is now generally available as a cloud-hosted service operated by Azure DevOps. Instead of deploying and maintaining an MCP server locally, developers can configure their AI clients to connect directly to a Microsoft-hosted endpoint.

Configuration requires only the Azure DevOps organization endpoint in the MCP configuration:

• Hosted endpoint: https://mcp.dev.azure.com/{organization}

• Transport: Streamable HTTP

• Authentication: Microsoft Entra

This approach significantly reduces setup complexity, allowing development teams to begin using AI-powered workflows without provisioning or maintaining additional infrastructure.

What Is the Azure DevOps MCP Server?

The Model Context Protocol (MCP) is an open protocol that allows AI assistants to securely interact with external applications and services using standardized interfaces.

In Azure DevOps, the MCP Server acts as a bridge between AI assistants and development projects. Once connected, AI tools can retrieve contextual information from Azure DevOps and use it to provide more relevant assistance during software development.

Depending on permissions granted to the user, AI assistants can work with resources such as:

• Work items

• Source code repositories

• Pull requests

• Azure Pipelines

• Project metadata

By providing richer project context, AI assistants can generate more accurate responses, automate repetitive tasks, and assist developers without requiring them to switch between multiple tools.

No Infrastructure to Manage

One of the primary advantages of the Remote MCP Server is that Microsoft hosts and manages the service.

Previously, organizations typically needed to deploy and maintain a local MCP server before AI assistants could access Azure DevOps resources. With the hosted service, developers simply configure their AI client to connect to the Azure DevOps endpoint.

This removes operational overhead while simplifying onboarding for development teams adopting AI-assisted workflows.

Authentication Through Microsoft Entra

The Remote MCP Server uses Microsoft Entra for authentication and authorization.

Because authentication relies on Microsoft Entra, Azure DevOps organizations must be backed by a Microsoft Entra tenant. Organizations that rely solely on Microsoft accounts (MSAs) are currently not supported.

Microsoft notes that this authentication model ensures AI assistants access only the Azure DevOps resources permitted by the authenticated user, aligning with existing enterprise security and identity management practices.

Supported AI Clients

Microsoft says several Microsoft and GitHub development tools already support the Remote MCP Server without additional onboarding.

Visual Studio Code with GitHub Copilot

Visual Studio Code users can connect GitHub Copilot directly to Azure DevOps projects. Once connected, Copilot gains contextual awareness of work items, repositories, pull requests, and pipelines, allowing it to deliver more relevant coding assistance and automate common development tasks from within the editor.

Microsoft AI Foundry

Developers building AI applications in Microsoft AI Foundry can connect Azure DevOps through the platform’s tools catalog. This enables AI agents and applications created in Foundry to interact with Azure DevOps resources as part of larger enterprise workflows.

Microsoft Copilot Studio

Microsoft has also added support for the Azure DevOps MCP Server in Microsoft Copilot Studio, its low-code platform for building custom AI agents.

Organizations can now create AI agents that securely access Azure DevOps data while integrating with Microsoft 365 and other enterprise services.

Additional Supported Clients

Other supported clients include:

• Visual Studio

• GitHub Copilot CLI

• GitHub Copilot app

These integrations expand the number of Microsoft development environments capable of using Azure DevOps project context through MCP.

Some Popular AI Clients Still Await Support

While the Remote MCP Server is generally available, not every AI client can connect today.

Microsoft explains that clients such as:

• ChatGPT

• Claude Desktop

• Claude Code

• Cursor

currently require additional Microsoft Entra capabilities, including support for dynamic OAuth client registration or client ID metadata documents, before they can authenticate with the hosted service.

Microsoft says it is working with the Microsoft Entra team to enable these authentication features. Until that work is complete, users of these AI clients should continue using the local Azure DevOps MCP Server.

This limitation is related to authentication compatibility rather than Azure DevOps functionality itself.

Local MCP Server Continues to Be Supported

Microsoft emphasized that the local Azure DevOps MCP Server is not being discontinued.

Organizations that cannot yet use the hosted service can continue running the local server while maintaining access to the same capabilities.

The company also confirmed it has recently aligned the toolsets between the local and remote implementations to maintain feature parity going forward. Future enhancements are expected to be delivered consistently across both deployment models.