Introducing the preview version of the VS Code AI Toolkit, now integrated with the MCP server.
VS Code AI Toolkit Preview: Now with MCP Server Integration
So, What Exactly Is the VS Code AI Toolkit?
It's a beefy extension that cuts the hassle out of building, testing, and deploying AI-powered stuff. You get out-of-the-box hookups to all the major players—OpenAI, Anthropic, Google—plus you can run models locally with ONNX and Ollama. Model fine-tuning, prompt engineering, evaluation... basically, it's your one-stop shop for AI development inside VS Code.
And now, with the new preview, it goes further: MCP server support lets you wire your AI agents straight into external APIs and services. Suddenly, your agent isn't just parroting text—it can actually do things.
What's New in This Release?
1. MCP Server Integration
This is the headline act. With the MCP server, you can:
-
Connect to external APIs and services: Want your agent to talk to a database? Or a web service? Or, honestly, whatever random app you’ve got running? Now it can.
-
Build your own MCP server: Templates in Python and TypeScript get you rolling, no matter how niche your needs.
-
Supercharge your agents: Grab live weather data, search your files, poke at cloud services—if you can build it, your agent can do it.
2. Agent Builder Now Supports MCP
The Agent Builder (formerly the "Prompt Builder") is MCP-aware now. That means you can:
-
Create complex prompts with structured outputs.
-
Chain prompts together for wild, multi-step workflows.
-
Plug in MCP tools for dynamic, real-world agent actions.
-
Test and debug prompts on the fly, with instant feedback.
3. Playground, But Upgraded
The AI Toolkit's Playground has gotten a glow-up:
-
Side-by-side model comparison, so you can watch different AIs duke it out in real time.
-
Multimodal input: text, images, file attachments—throw it all in.
-
Tweakable inference parameters like temperature and token limits, so you can fine-tune model behavior until it sings (or at least hums in tune).
4. Model Fine-tuning and Conversion
You can now fine-tune models locally or in the cloud, using GPU support via Azure Container Apps. Plus, the toolkit helps you:
-
Convert models from places like Hugging Face into ONNX for snappy local deployment.
-
Optimize models for CPU, GPU, or even NPU—pick your hardware poison.
Who Should Actually Use This Thing?
This update is for anyone who:
-
Builds apps: Want AI in your app, website, or mobile prototype? Go wild.
-
Works as an AI engineer: Fine-tune and deploy models on your own domain data.
-
Does data science: Compare models, run evaluations, tweak workflows.
-
Teaches or learns AI: Hands-on tools for classrooms, clubs, or late-night hacking.
Getting Started with MCP Servers
Ready to give MCP a spin?
-
Install the AI Toolkit from the Visual Studio Marketplace. Once it's in, you'll see its icon in your activity bar.
-
Head to Tools, hit + MCP Server to add one.
-
Pick your poison:
-
Featured MCP servers: Prebuilt for common tasks.
-
Connect to an existing MCP server: Already got one running? Just link it.
-
Build your own: Python or TypeScript templates are ready for you.
-
Once you’re set up, your MCP server shows up in the Agent Builder, ready to make your agents do more than just talk.
Why MCP Server Support Matters
Suddenly, your AI agents aren’t just stuck in the box—they can reach out and touch the real world. For example:
-
Dynamic data lookups: Agents can pull live stock prices or the current weather.
-
File wrangling: Automate local file searches, sorting, you name it.
-
Custom APIs: Hook up third-party services—payments, CRMs, whatever floats your boat.
Bottom line: your static AI agent just got a personality upgrade. Now it's a dynamic, multi-talented sidekick.
How to Try the Preview
Getting started is almost suspiciously easy:
-
Install the AI Toolkit for VS Code from the Marketplace.
-
Check out the Playground to mess with models and settings.
-
Use the Agent Builder to craft your own agents, with MCP servers in the mix.
-
Fine-tune models locally or in the cloud, however you like to roll.
Full details are in the AI Toolkit docs.
Final Thoughts
VS Code’s AI Toolkit, now with MCP server integration, takes building AI apps from “hmm, maybe someday” to “hey, let’s do it tonight.” Whether you’re hacking together a chatbot or architecting a whole smart system, all the tools are finally in one place.
If you’re serious about weaving generative AI into your next project, this preview is worth a shot. Don’t wait—grab the AI Toolkit today, and see how far your AI can actually go.
