AI is reshaping how we build, deploy, and scale software. As more apps become AI-powered, developers need environments that can keep up with speed and demands of innovation. We’ve heard from customers just how critical it is to have a zero-config experience when building AI applications, with streamlined access to compute, prebuilt models, and environment configuration tools.

We’ve been listening—and are excited to announce a new wave of capabilities in Microsoft Dev Box to accelerate AI development. Dev Box now empowers you to get started rapidly with ready-to-code cloud development environments optimized for AI workloads—whether you’re building copilots, training models, or iterating on multi-agent systems.

Here’s what’s new:

• Serverless GPU support: Spin up dev boxes with GPU acceleration — no additional setup required.
• Preconfigured access to Azure AI resources: Access Azure AI Services — directly from your dev box with secure identity passthrough.
• Agentic authoring workflow: Use natural language to generate and validate custom images for your team, powered by the Dev Box authoring agent.
• Dev Box MCP Server: Bring agent workflows into your IDE and CLI tools with the new Model Context Protocol server for Dev Box.

Let’s dive into the details.

Introducing zero-setup access to Serverless GPU compute in Microsoft Dev Box

As AI development becomes central to modern applications and services, developers increasingly need access to high-performance compute, and especially GPUs. But getting started with GPU workloads often means navigating complex infrastructure, provisioning the right SKUs, managing quotas, and configuring drivers. That’s a lot of friction between a developer’s idea and their first iteration.

Today, we’re making that process much simpler with serverless GPU support in Dev Box (Preview). With this new capability, developers can spin up GPU containers on demand directly from a dev box, with no infrastructure setup, manual configuration, or context switching.

Developer experience with serverless GPUs in Dev Box

For developers, there’s no need to provision VMs, manage containers, or handle drivers. Instead, they can launch a GPU-accelerated shell or VS Code session from their dev box in seconds.

The experience is zero-config by design:

• No VM creation or image management
• No manual authentication or network setup
• No GPU driver installation

Just launch and go—ideal for running inference tasks, accelerating model training, or experimenting with AI libraries.

Built-in governance for admins

While developers get an smooth, integrated experience, admins also get control and clarity. To make it easy for admins, serverless GPU compute is built into the existing Dev Box project structure, meaning admins can enable access per project and set usage limits without deploying or maintaining new infrastructure.

Admins can:

• Enable GPU compute for individual Dev Box projects
• Set and enforce limits on concurrent GPU sessions per project
• Monitor usage through familiar Dev Box management tools

This approach helps organizations empower teams to move fast, without compromising on governance, security, or cost control. Serverless GPU compute is currently available in West US 3, powered by NVIDIA T4 GPUs, and we will be expanding to additional regions and GPU SKUs in the coming months.

How to get started with serverless GPUs in Dev Box

Getting started is simple, whether you’re an admin enabling access or a developer jumping into a GPU session.

For Platform Admins: Enable GPU for a project

• Turn on the AFEC key to register and try out the Serverless GPU Preview
  • In the Azure Portal, navigate to the Subscription (in which the project is created), select ‘Preview features’, Register for ‘Dev Box Serverless GPU Preview’
• Navigate to the project and select “Dev Box Settings”
• Enable “Serverless GPU”
• Set the maximum number of concurrent GPU containers for the project

Once enabled, Dev Box users in that project will automatically see GPU options in their terminal and VS Code environments.

For Developers: Connect to a GPU with Zero Setup

After serverless GPU access is enabled, developers can connect using two easy methods:

Option 1: Launch a Dev Box GPU shell

• Open the Windows Terminal on your dev box
• Run: devbox gpu shell
• Instantly connect to a pre-configured GPU container and start coding, debugging

Option 2: Use VS Code with remote tunnels

• Open the Windows Terminal on your dev box
• Run: devbox gpu shell
• Launch Visual Studio Code
• Install the Remote Tunnels extension
• You will see an available tunnel named ‘gpu-session’, connect to it
• Start coding and debugging in a full-featured, GPU environment

Introducing preconfigured access to Azure AI resources

Many AI workflows don’t start with training a model from scratch. They start with adapting or fine-tuning existing models. Whether you’re building an agent, tuning prompts, or experimenting with inference, quick access to prebuilt LLMs can dramatically accelerate your development loop.

That’s why we’re excited to announce preconfigured access to Azure AI resources in Microsoft Dev Box (Preview). With just a few clicks, platform admins can enable access to Azure AI resources and developers can interact with large language models (LLMs) directly from their dev environment. No setup, no secrets management, and no context switching.

Develop faster with built-in LLM access

Once enabled, developers can access and deploy models from Azure AI Foundry using the Dev Box CLI. This streamlines key AI development workflows:

• Prompt tuning: Iterate on prompt engineering directly in your dev environment
• Model comparison: Evaluate different model deployments using live data
• Agentic workflows: Build and debug LLM-powered flows with real-time feedback

Now you can spend less time wiring up infrastructure and more time building.

How to get started with Azure AI Services in Dev Box

Whether you’re setting up access or deploying your first model, you can do this in a few simple steps.

For Platform Admins: Enable the resource called “Azure AI Services” for a Project

• Turn on the AFEC key to register and try out the Azure AI Services Preview
  • In the Azure Portal, navigate to the Subscription (in which the project is created), select ‘Preview features’, Register for ‘Dev Box Azure AI Services Preview’
• Navigate to your Dev Box project, go to Settings > Dev Box Settings
• Toggle “Azure AI Services” to Enabled
• Click Apply

This creates a managed Azure AI Services instance paired with the project, hosted in a resource group managed by Dev Center.

For Developers: Manage and use LLMs with the Dev Box CLI

After the feature is enabled, developers are automatically granted access to the Azure AI instance with the Cognitive Services User role. Here’s how to start using it in the command line:

• Check your access: ‘devbox ai status’ to confirm that your dev box is connected to the AI Services instance.
• List available models: ‘devbox ai list-models’ to browse the catalog of models available in your project.
• Deploy a model: ‘devbox ai deploy-model –model-name <model-name> [–deployment-name <deployment-name>]’ to deploy an instance of a model for use in your application.
• View deployed models: ‘devbox ai list-deployments’ to get the endpoint and deployment name needed to integrate the model into your app.
• Launch Azure Foundry portal: ‘devbox ai start-ai-foundry-portal’ to open the Foundry portal to manage deployments and explore model options via a visual UI. We recommend using Azure OpenAI SDK with Entra ID authentication instead of API key auth for better security and enterprise compliance.

Want to try it out quickly? Clone the OpenAI SDK Tutorial and configure the appsettings.json file with your model’s deployment details. Then hit F5, sign in using Entra ID, and you’re up and running with a fully functional LLM-powered app.

Customize your Dev Box configurations with the new agentic workflow

Project onboarding is still one of the biggest productivity bottlenecks in software development, especially for teams building complex, AI-powered applications. Before writing a single line of code, developers often spend hours wrangling toolchains, resolving dependency conflicts, and aligning environment settings.

At Ignite 2024, we introduced team customizations and built-in imaging functionality for Dev Box, which allows platform engineers and dev leads to define a reusable image configuration using a single file: imagedefinition.yaml. This ensures that every developer on the team gets a consistent, ready-to-code environment without the manual setup.

The response has been overwhelmingly positive, but we also heard your feedback loud and clear: Authoring these YAML files manually can be slow, error-prone, and not fun. We agree. YAML should empower—not intimidate. So we asked: “What if Dev Box could help generate these files for you — based on your intent, your repo, or even your current machine?”

Meet the Dev Box authoring agent (Preview)

We’re excited to introduce the authoring agent, an AI-powered, conversational workflow built into the Dev Box extension for Visual Studio Code.

With the authoring agent, platform engineers and dev leads can generate and refine imagedefinition.yaml files through a natural, chat-based experience with full transparency and control. No need to memorize schemas, hunt for allowed packages, or worry about indentation quirks.

The authoring agent workflow supports three core scenarios:

• Mimic your current machine: Capture your local dev setup and generate an equivalent image definition.
• Use repository context: Point the agent at a repository and it will generate a recommended imagedefinition.yaml based on the tech stack, build files, and dependencies it finds.
• Natural language instructions: Describe what you need — e.g., “Set up a Python 3.12 environment with VS Code and PostgreSQL” — and let the agent create the definition for you.

How to get started with the Dev Box authoring agent

• Open Visual Studio Code: Make sure you’re on VS Code version 1.99 or later.
• Install the latest Dev Box extension: Open Extensions (Ctrl+Shift+X), search for Dev Box, and install or update the extension.

• Enable Agent Mode: In the Settings editor, enable the agent.enabledsetting.
• Launch Copilot Chat:
  • Open Copilot Chat panel
  • Under Select tools, make sure Dev Box tools are pre-selected

1. • Select Agent Mode and Choose the model: Claude 3.5 Sonnet

• Provide natural language prompts, such as:
  • “I want to configure a dev box with all the tools and packages required to work on this [repo name] repo.”
  • “I want to pre-configure a dev box with Visual Studio 2022 Enterprise, VS Code, Git, .NET SDK 8, Node.js LTS, Docker Desktop installed, and have the team’s repo [URL] cloned onto the dev box.”
  • “I want to configure a dev box with all the dev tools and packages installed on my current machine.”

Tip: Clone and open the specific repo in VS Code if you want to generate the definition in the context of a repository

• Follow prompts to configure packages
  • When prompted, select Continue to proceed with package configuration
  • Copilot will generate the imagedefinition.yaml file
• Refine with additional prompts
  • Continue interacting with the agent until your desired tools and packages are reflected in the file

Validating or applying the customizations

These steps must be performed within a Dev Box instance.

• Select Continue when prompted to proceed with validation or provide the prompt to validate the imagedefinition.yaml
  • Submit a prompt to the agent: “Validate my imagedefinition.yaml file.”
• Apply customizations on the current Dev Box
  • Open Command Palette (Ctrl+Shift+P)
  • Select Apply Customization Tasks
  • Confirm the UAC prompt to install tools and apply settings

Save and configure the project to leverage the image definition

Once your imagedefinition.yaml is ready:

• Save the file in a GitHub or Azure DevOps repository.
• Attach the repository as a catalog to your Project.
• Configure a Dev Box pool using the generated imagedefinition.yaml:
  • Go to Dev Box Pools in your Project.
  • Create new or edit an existing dev box pool.
  • Select the image definition created with your imagedefinition.yaml.

From there, every Dev Box created in the pool will launch with a consistent, preconfigured environment tailored to your project needs.

The Dev Box authoring agent represents a leap forward in dev environment automation — reducing friction for platform engineers and dev leads and accelerating onboarding for developers. It’s a great example of how agentic workflows and AI-native tooling can transform software development at scale.

Introducing the Dev Box MCP: AI-native control for your development environment

The way developers work is rapidly evolving. AI agents and copilots are increasingly handling operational tasks—from spinning up dev environments to retrieving workspace context—allowing developers to focus on building great software.

At the center of this shift is the Model Context Protocol (MCP), an emerging standard that enables agents to discover and invoke tools programmatically. Today, we’re excited to bring Dev Box into this new ecosystem.

Meet the Dev Box MCP: Open Source and agent-ready

We’re announcing the preview of the Dev Box MCP, an open-source server that exposes Dev Box as a programmable resource through the MCP standard. This makes it easier for both developers and AI agents to automate Dev Box actions like provisioning, starting, stopping, and listing environments — all via structured, AI-friendly interfaces.

Whether you’re using an AI assistant, a CLI, or building your own developer tool, the Dev Box MCP gives you direct control over your environments without ever opening a portal.

How to get started with Dev Box MCP Server

The Dev Box MCP Server will soon be published as a public NPM package. Once available, you’ll be able to install and launch it using any MCP-compatible client. Follow the below instructions to start using Dev Box MCP Server today.

Sample MCP Client Setup (VS Code)

To leverage the MCP server for Dev Box within VS Code, the following JSON can be added to the users’ MCP settings:

    "mcp": {
        "servers": {
            "Dev Box MCP": {
                "type": "stdio",
                "command": "npx",
                "args": [
                    "-y",
                    "https://aka.ms/devbox/mcp/preview"
                ]
            }
        }
    }

Once the server is added, start it through your MCP-compatible client. Upon successful startup, you’ll have access to 31 Dev Box tools — covering everything from provisioning to repair to snapshot management.

With the server running, you can now use natural language to invoke Dev Box operations directly from GitHub Copilot Agent mode. Try these sample prompts (and more):

• Create a dev box for me: This experience will guide you through all the available projects and pools you have access to, allow you to create an appropriate dev box for your need
• Skip shutdown on my dev box – <name of dev box>: This experience will walk through the journey of skipping shutdown on an already available dev box. It can also ensure that there are no pending actions already scheduled.
• Install .NET9, Visual Studio and VSCode within my dev box – <>: Using this command, you can personalize your machine with available tools, which can be installed remotely, without any user interference.
• Change my dev box theme to dark mode: Personalize your dev box without logging in, directly from your client.

No UI clicks, no dropdowns, just expressive calls using tools that speak MCP.

What’s Next

This release marks the beginning of a long-term vision to make Dev Box environments fully accessible and controllable through AI-native workflows.

We’re actively working on:

• Expanding action coverage: Supporting more advanced Dev Box operations and richer project interactions
• Deep ecosystem integrations: Enabling seamless use with popular AI agents, terminals, IDEs, and workflows
• Enhanced usability for assistants: Making it easier for LLMs to reason about, validate, and invoke Dev Box APIs
• Broader AI automation: Supporting complex, multi-step flows for LLM-driven software engineering

We believe Dev Box MCP will be a foundational building block for the future of cloud-based development, putting structured, programmable environment control into the hands of every developer — and every agent.

Transform your development experience with Microsoft Dev Box

As AI-driven software development accelerates, developers need environments that are fast, flexible, and secure — environments that evolve with the way they work.

Our mission is simple: make Microsoft Dev Box the preferred development environment for enterprise developers by delivering self-service, ready-to-code setups, and AI-powered workflows that unlock speed and innovation across teams.

With new capabilities that support AI development, enable agentic automation, and streamline environment customization, Dev Box empowers your teams to move faster and focus on what matters — building great software.

Ready to transform how your teams build?

• Watch our breakout session: Unleash developer potential with AI and Dev Box
• Explore what’s coming next: Check out our public roadmap
• Get started today: Try Microsoft Dev Box

We’re continuing to make Dev Box even better for you — so please keep the feedback coming. We can’t wait to see what you’ll build!