AI Toolkit for VS Code just hit a major milestone with its February 2026 v0.30.0 release, and Microsoft is turning VS Code into a serious, end‑to‑end environment for building production‑grade AI agents. The latest update focuses on three big themes: making tools easier to discover, turning agent debugging into a first‑class workflow, and treating evaluations like real tests that fit naturally into CI and DevOps pipelines.
AI agents, without leaving VS Code
In its latest Microsoft Developer Community Blog post, Microsoft outlines how the AI Toolkit for VS Code — February 2026 Update brings the extension to version 0.30.0 and pushes it firmly into “idea → code → debug → evaluate → deploy” territory, all inside the editor. The release reframes the Toolkit as more than a collection of AI helpers: it is now a coherent environment for discovering tools, wiring up agents, debugging multi‑agent workflows, and running structured evaluations that feed into Microsoft Foundry.
At a high level, the update targets three pain points developers have struggled with when building agents: scattered tool configurations, opaque runtime behavior, and ad‑hoc evaluation scripts that are hard to maintain. Version 0.30.0 addresses each of these with the new Tool Catalog, the Agent Inspector, and Evaluation as Tests, while layering on usability and performance improvements across Agent Builder, Model Catalog, Copilot‑powered workflows, and profiling.
Tool Catalog: one hub for every agent tool
The new Tool Catalog is the centerpiece of the v0.30.0 release, acting as a centralized hub where developers can discover, configure, and manage tools that their agents rely on. Instead of jumping between JSON files, environment variables, and scattered configuration snippets, you get a single, unified surface in VS Code to see everything your agents can call.
From this Catalog, you can:
-
Browse, search, and filter tools coming from the public Microsoft Foundry catalog and local studio MCP servers.
-
Configure connection settings for each tool directly in VS Code, so authentication and endpoints live alongside your code.
-
Add tools to agents seamlessly using Agent Builder, without manually editing configuration files.
-
Manage the full lifecycle—adding, updating, and removing tools as your agent’s capabilities evolve.
The practical impact is that expanding an agent’s capabilities goes from “edit three different configs and hope nothing breaks” to “a few clicks in a dedicated UI.” For teams building larger, multi‑tool agents on top of Foundry, that centralization should make day‑to‑day maintenance less fragile and more transparent.
Agent Inspector: debugging agents like real software
The new Agent Inspector is Microsoft’s answer to the “black box” feeling many developers have when working with complex AI agents. In v0.30.0, you can press F5 to launch your agent with full debugger support, turning agent runs into something that looks and feels like debugging any other application inside VS Code.
Featured capabilities include:
-
One‑click F5 debugging with breakpoints, variable inspection, and step‑through execution for agent code.
-
Copilot‑driven auto‑configuration that scaffolds agent code, endpoints, and debugging setup, so you spend less time wiring and more time iterating.
-
Generation of production‑ready code via the Hosted Agent SDK, designed to deploy into Microsoft Foundry.
-
Real‑time visualization of streaming responses, tool calls, and multi‑agent workflows so you can see which parts of your system are doing what, and when.
-
Quick navigation from workflow nodes straight into source files by double‑clicking.
-
A unified view that blends chat interactions with workflow visualization, keeping conversation context and execution traces in the same place.
This moves agent development closer to traditional app development, where you expect to pause execution, inspect state, and replay scenarios instead of relying purely on logs or trial‑and‑error. For teams trying to ship agents into production on Foundry, the ability to debug with that level of fidelity directly in VS Code should make it easier to track down subtle issues and understand emergent behavior.
Evaluation as Tests: quality treated like code
Another headline feature in v0.30.0 is “Evaluation as Tests,” which reimagines evaluation as something you define, run, and version like any other test suite. Rather than scattered evaluation scripts, the Toolkit now plugs evaluations into familiar developer workflows centered on pytest and the VS Code Test Explorer.
Here’s what this looks like in practice:
-
You define evaluations as test cases using pytest syntax, enhanced with annotations from the Eval Runner SDK.
-
Those evaluations can be executed directly from the VS Code Test Explorer, alongside unit and integration tests, mixing and matching scenarios as needed.
-
Results are surfaced in a tabular view, with Data Wrangler integration to help you slice, filter, and analyze outcomes in more depth.
-
Once an evaluation definition is ready, you can submit it to run at scale in Microsoft Foundry, bringing the same scenarios into your hosted environment.
By anchoring evaluations in test frameworks and CI‑friendly workflows, Microsoft is positioning agent quality as something you can track over time rather than as one‑off experiments. For organizations standardizing on Foundry, having the same evaluation definitions run both locally and in the cloud should help close the gap between development and production.
Agent Builder, Model Catalog, and Copilot get smarter
The February 2026 update doesn’t stop at flagship features; it also delivers a set of quality‑of‑life improvements across the rest of the Toolkit. Agent Builder, Model Catalog, and GitHub Copilot‑powered workflows all see enhancements aimed at making routine tasks faster and more intuitive.
On the Agent Builder side, v0.30.0 introduces:
-
A redesigned layout that improves navigation and puts key elements in more logical places.
-
A quick switcher to move between agents without losing your place.
-
Support for authoring, running, and saving Foundry prompt agents directly from the Toolkit.
-
The ability to attach tools from the Tool Catalog or built‑in tool set to those Foundry prompt agents.
-
A new “Inspire Me” feature to help you draft agent instructions when you are not sure how to phrase behavior or capabilities.
-
Broad performance and stability improvements that should make day‑to‑day use feel snappier.
In the Model Catalog, Microsoft adds support for models that expose the OpenAI Response API, including the gpt‑5.2‑codex coding‑optimized model. That means you can bring in newer API‑style models and integrate them into your workflows via the same unified catalog. The update also includes general performance and reliability tweaks so browsing and selecting models feels more responsive.
The “Build Agent with GitHub Copilot” experience also expands with a new Workflow entry point that helps you quickly generate multi‑agent workflows with Copilot‑assisted scaffolding. You can orchestrate these workflows by selecting prompt agents from Foundry, which makes it easier to go from a conceptual multi‑agent design to runnable code.
Conversion, profiling, and hardware‑aware tuning
Beyond the core IDE experience, v0.30.0 brings deeper profiling and conversion tools aimed at teams pushing agents into more specialized environments. Developers can now generate interactive playgrounds for history models, which helps when you need to test and explore models that are tuned to conversation history or multi‑turn interactions.
The Toolkit also adds Qualcomm GPU recipes, signaling a continued focus on running AI workloads efficiently across a broader hardware landscape. For Microsoft’s own Phi Silica model, the update now surfaces resource usage directly in the Model Playground, giving developers clearer visibility into how much compute a given configuration consumes. That information can guide decisions about deployment targets, scaling, and cost‑optimization strategies.
Why this release matters
Microsoft has been steadily transforming AI Toolkit for VS Code from a convenient add‑on into a full agent‑development environment, and the February 2026 v0.30.0 release is one of the clearest inflection points so far. With the Tool Catalog, Agent Inspector, and Evaluation as Tests, the extension now addresses discovery, debuggability, and quality in a way that aligns with how professional software teams already work.
For developers keeping an eye on Microsoft’s broader web platform push, including its latest Edge Interop 2026 efforts, we’ve covered how the browser is doubling down on real‑world compatibility issues.
For developers building on Microsoft Foundry, the tighter loop between AI Toolkit for VS Code and the hosted platform—spanning tool management, Hosted Agent SDK code generation, and scalable evaluations—should make it more realistic to treat agents as first‑class production software, not just experiments. And for the broader VS Code ecosystem, this update reinforces the editor’s role as a central hub for AI‑driven development, from local prototyping all the way to cloud deployment.
Recent Posts You Might Like
- Brilliant VS Code Becomes First AI Code Editor to Give MCP Agents a Visual Voice
- How to Access GPT‑5.2 in Microsoft 365 Copilot Chat (Boost Quick Response vs Smarter Think Deeper)
- Claude Opus 4.6 Brings Next‑Gen Agentic Coding to GitHub Copilot Users Now Available
- SQLCon 2026: Six Reasons You Can’t Miss This Epic Event in Atlanta
- Microsoft Launches Azure Static Web Apps Skill for GitHub Copilot, Turning AI Into a Deployment Sidekick
Discover more from Microsoft News Now
Subscribe to get the latest posts sent to your email.
