Microsoft Foundry Dev Pack Ships as One Install for Agent Builders
The pack bundles the Foundry CLI, a skill for coding agents, the VS Code toolkit and a GitHub Copilot app plugin, all generally available now.
One install, four places to work
On November 18, 2026, Microsoft said the Microsoft Foundry Dev Pack is generally available, announced at Microsoft Ignite. The pack puts the company's Foundry developer tools onto a developer's machine in a single install, covering scaffolding, debugging, evaluating and deploying agents from an editor, a terminal or a coding agent. Microsoft describes it as the recommended way to set up a machine for Foundry, repeatably. The full details are in the original Foundry Dev Pack announcement.
What one install actually adds
The Dev Pack combines two tools, the Foundry CLI (azd ai) and the Microsoft Foundry Skill. It also installs the Foundry Toolkit for VS Code when VS Code is present, and the Foundry plugin for the GitHub Copilot app when that app is installed. Microsoft says the pack builds on the Foundry CLI reaching general availability on November 5.
The tools share project files, so a project can be scaffolded in one surface, debugged in another and deployed from whichever fits. Microsoft's install commands are:
- Windows:
winget install Microsoft.FoundryDevPack - macOS:
brew install --cask microsoft/foundry/devpack && foundry-devpack install - Linux:
curl -fsSL https://aka.ms/foundry-devpack-install.sh | bash
The post says those commands can go into onboarding docs, dev containers and machine setup scripts so a team starts from the same toolset.
What each surface does
The Foundry Toolkit for VS Code is the fullest set: browse Foundry project resources, try models in the Model Playground, scaffold an agent and press F5 to debug, inspect requests and responses in Agent Inspector, and keep models, agent code, resources, tests and evaluations in the repository. Traces point at a problem, and redeployment happens from the editor.
The Foundry CLI turns agent work into repeatable commands and project configuration in azure.yaml, taking the same workflow from a laptop into scripts and CI/CD pipelines, with infrastructure customisable through Bicep or Terraform. Target environments and cost-bearing actions stay explicit and reviewable, per the post.
In the GitHub Copilot app, the Foundry plugin opens Microsoft Foundry Canvas, a guided, project-aware experience beside the chat. You pick the agent's model, tools, skills and guardrails, Copilot generates and updates the project files, and you inspect and test locally. Built-in solution templates are offered as starting points.
The Microsoft Foundry Skill carries reusable workflows for creating, deploying, evaluating and troubleshooting Foundry agents across compatible hosts, which the post lists as GitHub Copilot, Claude Code, Cursor and other coding-agent hosts. It guides an agent to use tools such as az, azd, MCP tools and repository scripts, and Microsoft says this helps agents skip broad web searches and repeated trial and error, with fewer tokens and faster iteration. The developer reviews plans, file changes, commands and cost-bearing actions before they run.
The terminal loop, step by step
The post walks through a Python Agent Framework sample. Before starting it lists Azure Developer CLI 1.37.0 or later, a current uv release and Python 3.13 or later as requirements. Then:
azd auth login
azd ai agent init --template <sample azure.yaml URL> --deploy-mode code --src .
azd provision
azd ai agent run
azd ai agent invoke --local "Write a haiku about deploying cloud applications."
azd deploy
After that, azd ai agent monitor --follow streams logs, azd ai agent doctor diagnoses setup and permission issues, and azd pipeline config sets up GitHub Actions or Azure DevOps.
Limits and open questions
The post gives no pricing for any of the tools, and does not say which hosts beyond the three it names count as compatible for the Skill. It says the GitHub Copilot app plugin installs when that app is present but does not cover the app's own availability. Foundry Canvas is described only at the level of choosing model, tools, skills and guardrails. The version requirements are attached to the Python sample, not to every language. That the four surfaces share project files cleanly is stated rather than shown.
What this likely changes for builders
This reads less like a new runtime and more like an attack on setup friction. The Skill is likely the lowest-effort entry point for teams already working in Cursor or Claude Code, while the VS Code Toolkit carries the deeper feature set for people who live in an editor. Teams that bake the install commands into dev containers probably get a more consistent environment than ad hoc setups, though the post does not make that comparison itself.
What it means for building apps with AI
If you are building an app with an assistant rather than running an agent pipeline, the takeaway is narrower: agent tooling is converging on skills and shared project files instead of one IDE. Draftship takes a different route, where you describe an app in plain language and it writes, builds and hosts the code with a built-in database and sign-in. If you are wiring model calls into something you have already shipped, see adding AI to your app. Draftship is not affiliated with Microsoft and makes no claim about Foundry tooling.
Primary sources
- The Foundry Dev Pack is here: build Foundry agents your way · Microsoft Foundry Blog, Nov 18, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.