AzureFeeds

February brings a major milestone for AI Toolkit. Version 0.30.0 is packed with new capabilities that make agent development more discoverable, debuggable, and production-ready—from a brand-new Tool Catalog, to an end-to-end Agent Inspector, to treating evaluations as first-class tests.

🔧 New in v0.30.0

🧰 Tool Catalog: One place to discover and manage agent tools

The new Tool Catalog is a centralized hub for discovering, configuring, and integrating tools into your AI agents.

Instead of juggling scattered configs and definitions, you now get a unified experience for tool management:

• Browse, search, and filter tools from the public Foundry catalog and local stdio MCP servers


• Configure connection settings for each tool directly in VS Code


• Add tools to agents seamlessly via Agent Builder


• Manage the full tool lifecycle: add, update, or remove tools with confidence

Why it matters: expanding your agent’s capabilities is now a few clicks away—and stays manageable as your agent grows.

🕵️ Agent Inspector: Debug agents like real software

The new Agent Inspector turns agent debugging into a first-class experience inside VS Code. Just press F5 and launch your agent with full debugger support.

Key highlights:

• One-click F5 debugging with breakpoints, variable inspection, and step-through execution


• Copilot auto-configuration that scaffolds agent code, endpoints, and debugging setup


• Production-ready code generated using the Hosted Agent SDK, ready for Microsoft Foundry


• Real-time visualization of streaming responses, tool calls, and multi-agent workflows


• Quick code navigation—double-click workflow nodes to jump straight to source


• Unified experience combining chat and workflow visualization in one view

Why it matters: agents are no longer black boxes—you can see exactly what’s happening, when, and why.

🧪 Evaluation as Tests: Treat quality like code

With Evaluation as Tests, agent quality checks now fit naturally into existing developer workflows.

What’s new:

• Define evaluations as test cases using familiar pytest syntax and Eval Runner SDK annotations


• Run evaluations directly from VS Code Test Explorer, mixing and matching test cases


• Analyze results in a tabular view with Data Wrangler integration


• Submit evaluation definitions to run at scale in Microsoft Foundry

Why it matters: evaluations are no longer ad-hoc scripts—they’re versioned, repeatable, and CI-friendly.

🔄 Improvements across the Toolkit

🧱 Agent Builder

Agent Builder received a major usability refresh:

• Redesigned layout for better navigation and focus


• Quick switcher to move between agents effortlessly


• Support for authoring, running, and saving Foundry prompt agents


• Add tools to Foundry prompt agents directly from the Tool Catalog or built-in tools


• New Inspire Me feature to help you get started when drafting agent instructions


• Numerous performance and stability improvements

🤖 Model Catalog

• Added support for models using the OpenAI Response API, including gpt-5.2-codex


• General performance and reliability improvements

🧠 Build Agent with GitHub Copilot

• New Workflow entry point to quickly generate multi-agent workflows with Copilot


• Ability to orchestrate workflows by selecting prompt agents from Foundry

🔁 Conversion & Profiling

• Generate interactive playgrounds for history models


• Added Qualcomm GPU recipes


• Show resource usage for Phi Silica directly in Model Playground

✨ Wrapping up

Version 0.30.0 is a big step forward for AI Toolkit. With better discoverability, real debugging, structured evaluation, and deeper Foundry integration, building AI agents in VS Code now feels much closer to building production software.

As always, we’d love your feedback—keep it coming, and happy agent building! 🚀