"Stop using AI," reads yet another viral post. I get it. It's frustrating to review a colleague's auto-generated work, filled with AI's classic giveaways like generic code comments and phrases like "In today's fast-paced world..."

Still, AI plays a pivotal role in my career. I don't rely on AI to do my work, but I use it to help me brainstorm and work more effciently. The introduction of Model Context Protocol (MCP) has made this even easier. MCP is an open standard that gives AI tools the context they need to be useful in the real world. It enables AI agents to interact with APIs, apps, and systems in a structured way. I use Codename goose, an open source AI agent built on MCP.

Here are 11 real ways I use AI Agents without sacrificing authenticity, creativity, or quality:

We brought Goose to New York City and it was one for the books.

Over 100 people registered, plus 87 on the waitlist, and we had a packed room full of folks who were curious, thoughtful, and ready to dive in. Some were developers already exploring Goose and MCP, others were totally new to the world of AI agents. That’s the beauty of Goose, it’s for developers and non-developers.

The energy was there from the moment the event began - music, pizza, and authentic networking. We had lightning talks, a Goose-themed game, hands-on hacking, and yeah… a few Ebbs IPAs might’ve ended up in people’s backpacks by the end of the night.

As the goose team continues to work on new features over 800 community members voted to express their need for Goose on Windows. Today's top contributor stepped in to aid this ongoing effort.

Our recently published Goose benchmark revealed significant performance limitations in models where tool calling is not straightforwardly supported (e.g., Gemma3, Deepseek-r1, phi4). These models often fail to invoke tools at appropriate times or produce malformed or inconsistently formatted tool calls. With the most recent releases of Llama4 and Deepseek v3 (0324), we are again observing challenges with effective tool calling performance, even on these flagship openweight models.

You ever open a GitHub repo or blog post, read the first sentence, and immediately feel like you’ve stumbled into a PhD dissertation?

Yeah. Same.

MCP (Model Context Protocol) sounds complicated, but it’s really not. Think of this as your go to cheat sheet, no whitepapers, no academic jargon, just plain English and a few good visuals.

On Feb 2, 2025, Andrej Karpathy coined the phrase "vibe coding". Vibe coding represents a new approach to coding where developers ask an AI agent to build something, and they go with the flow.

The Model Context Protocol (MCP) makes this practice possible. Before MCP, developers copied and pasted context between applications. This workflow fell short of the promised AI agent automation that everyone claimed. Today, AI agents can work autonomously using MCP and integrate with any application, from GitHub to Cloudflare, YouTube, and Figma.

This shift democratizes coding. For example, it's empowered:

• Web developers to create video games with Unity
• Designers and product managers to prototype full-stack applications
• Business owners to transform their visions into functional products

It's a freeing experience. But too often, we're Icarus with the keyboard, vibe coding too close to the sun.

As a developer, finding the right tools that seamlessly work together can feel like discovering a superpower. And when you have a working process, it can sometimes be difficult to try out new tools.

With the introduction of MCPs, AI agents like Goose are able to plug in to my existing tools, and the only thing that changes with my workflow is that much welcomed automation that comes with it. I still do the same things I do, but backed by AI, I can now do them faster and with more confidence.

Today, I'm excited to share not just my favorite MCP servers, but the ones I actually use almost everyday with real applications that you can probably relate to as well.

MCP this, MCP that, what exactly is it, and can you use them if you're not a developer? 🤔

We've been measuring Goose's performance with various AI models, including a variety of popular open-source models that can run locally on consumer hardware (RTX 4080, Mac M-series). We understand that many in our community value a fully open-source, local experience without relying on cloud services.

This blog shares our findings comparing open-source models against their closed counterparts, highlighting both current performance gaps and paths for future improvement. Our benchmark is still in its early stages, but we wanted to release it as a starting point for distinguishing models that exhibit stronger agentic capabilities by their ability to pilot Goose (distinct from reasoning or other capabilities often captured in other popular benchmarks).

Authors: Alex Rosenzweig, Arihant Virulkar, Andrea Leoszko, Wes Ring, Mike Shema, F G, Alex Klyubin, Michael Rand, Zhen Lian, Angie Jones, Douwe Osinga, Mic Neale, Bradley Axen, Gelareh Taban

At Block, we’ve been working hard to augment the capabilities of AI tooling by building "MCP Servers" which are designed to help make our Artificial Intelligence (AI) Agent codename goose more capable of interacting with the systems and tools we care about.

Block’s Information Security (InfoSec) team has been heavily involved in this work and we wanted to capture our learnings in the space to help others. We expect there to be growing adoption and use cases for this including applying the technology in the security domain.