Anthropic's New Model Context Protocol in 10 Minutes

Just today, Anthropic announced that they're open sourcing what they're calling Model Context Protocol, or MCP for short. They described this as a new standard for connecting AI assistants to the systems where data resides, including content repositories, business tools, and development environments. The goal is to help frontier models produce better, more relevant responses. In this video, I'll read through the blog post and then point you in the direction of how you can get started with integrating this into your applications as AI assistants gain mainstream adoption.

The industry has invested heavily in model capabilities, achieving rapid advances in reasoning and quality. Yet, even the most sophisticated models are constrained by their isolation from data trapped behind information silos and legacy systems. Every new data source requires its own custom implementation, making truly connected systems difficult to scale. MCP addresses this challenge. It provides a universal open-ended standard for connecting AI systems with data sources, replacing fragmented integrations with a single protocol.

The result is simpler, more reliable ways to give AI systems access to the data they need. They then go further to describe the Model Context Protocol. In terms of some of the specifics of the Model Context Protocol, it is described as an open-source standard that enables developers to build secure two-way connections between their data sources and AI-powered tools. The architecture is straightforward: developers can either expose their data through MCP servers or build AI applications as MCP clients that connect to these servers.

Today, they're introducing three major components of the Model Context Protocol for developers. For the Model Context Protocol, they have SDKs available in both TypeScript and Python. You are able to access the local MCP server through the Claude desktop app at the time of recording. Additionally, there is an open-source repository of MCP servers that you can explore.

They described that Cloud 3.5 Sonnet is adept at quickly building MCP servers, making it easy for organizations and individuals to rapidly connect their most important data sets with a range of AI-powered tools. To help developers start exploring, we're sharing pre-built MCP servers for popular enterprise systems like Google Drive, Slack, PostgreSQL, and Puppeteer. Early adopters like Block and Apollo have integrated MCP into their systems, while development tool companies including Zed, Replit, Codium, and Sourcegraph are working with MCP to enhance their platforms. This collaboration enables AI agents to better retrieve relevant information, further understanding the context around coding tasks and producing functional code with fewer attempts. At Block, open source is more than just a development model; it's a foundation of our work and our commitment to creating technology that drives meaningful change and serves as a public good for all.

Open technologies, like the Model Context Provider, are the bridge connecting AI to real-world applications, ensuring that innovation is accessible, transparent, and rooted in collaboration. We are excited to partner on a protocol and use it to build agentic systems, which remove the burden of the mechanical so people can focus on the creative. Furthermore, they described that instead of maintaining a separate connector for each data source, developers can now build against a standard protocol as the ecosystem matures. AI systems will maintain context as they move between tools and data sets, replacing today's fragmented integrations with a more sustainable architecture. Developers can start building and testing with MCP connectors today.

Existing Cloud for Work customers can begin testing MCP servers locally, connecting the cloud to internal systems and data sets. We'll soon provide a developer toolkit for deploying remote production MCP servers that serve your entire Cloud for Work organization. To get started, you can install the pre-built MCP servers through the Cloud desktop app. You can follow the quick start guide to build your first MCP server. To contribute to the open-source repositories of connectors and implementations, you can do so through the link, which I'll also include in the description of the video.

They describe that we're committed to building MCP as a collaborative open-source project and ecosystem. We're eager to hear your feedback, whether you're an AI tool developer, an enterprise looking to leverage existing data, or an early adopter exploring the frontier. We invite you to build the future of context-aware AI together. Now, in terms of the protocol itself, you can find this on GitHub at github.com/model-context-protocol. Within this, there are example specifications as well as discussions. Let's just take a look at one of the SDKs, for instance, the TypeScript SDK. There is an overview that allows applications to provide context for LLMs in a standardized way, separating the concerns of providing context from the actual LLM interaction with the SDK.

This is what you're going to use to build these "quote unquote" MCP clients that you can connect to MCP servers. What the SDK does is allow you to build MCP clients that can connect to any MCP server, create MCP servers that expose resources, prompts, and tools using standard training. It also handles all MCP protocol messages as lifecycle events. To get started, you can simply install it. If we take a quick look at the quick start, it appears relatively simple to begin.

You can create the path to your server. You can create a new client. You can connect to that transport protocol, and then you can list out all of the various resources. What's great about this is that, if you just want to expose to the model what it has access to and what it's capable of, you'll be able to do this in a more general way. At a glance, this looks like it will be able to give the models effectively the context of knowing what they are able to accomplish through this MCP protocol.

It can access things like Puppeteer. It will be able to access your PostgreSQL server and handle all of those different interactions. This video is brought to you by Scrimba, the innovative coding platform that brings interactive learning to life. Dive into a variety of courses from AI engineering to front-end development, Python, UI design, and much more. Scrimba is game-changing.

Their unique scrim screencast format allows you to pause the lesson at any time and start directly editing the teacher's code. Their curriculum is built in collaboration with industry leaders including Mozilla, MDN, Hugging Face, and LineChain. It includes building applications with OpenAI's cloud Mistral models and guides you on deploying projects to platforms like Cloudflare. While AI tools can assist with coding, a solid grasp of the fundamentals is essential for gaining real experience. Scrimba offers something for everyone, from complete beginners to advanced developers, and about 80% of Scrimba's content is completely free. Sign up for a free account today using my link below and enjoy an extra 20% discount on their pro plans when you're ready to upgrade. I'm sure you'll love it!

Now, what's going to be interesting about this is how it actually operates and performs with other models within this environment. Let's take a look at the Puppeteer browser automation, for instance. Here's an example of Puppeteer, which is a synthetic browser that many people use for web scraping and related tasks. You can take screenshots, navigate web pages, execute JavaScript, and more. Here’s an example of a server that has a variety of different tools.

You can use Puppeteer to navigate, take screenshots, click, fill out forms, and basically utilize all of the core components that are built within the Puppeteer library. If you have an AI agent that requires the capability to navigate the web, you could connect it to one of these types of servers to provide that functionality. Here's a brief deep dive into this. You have the specified tools here; we see that the server is being defined, and then we are setting up all of the tools that Puppeteer has available for leverage.

You can navigate to a URL, take a screenshot, click on the page, and fill out an input within the page. You can also evaluate the JavaScript on the page. Here is an example of the logic for each different function. We click what looks like the screenshot functionality, and essentially, all of the other cases are contained within this. Effectively, how this works is that you have the ability to list all the resources available on your server.

It will be able to tell you, "Okay, this is a Puppeteer server. It can take screenshots, it can perform actions, it can navigate to URLs, and you can actually invoke those actions." You can call the requested tools that the LLM model you are using understands, and it will have the ability to recognize what the server can do and utilize that server effectively. It's really interesting because, all of a sudden, the LLM can call these servers and understand their capabilities.

And depending on the task or the query, it can perform those actions. Now, what's interesting about this is it doesn't necessarily just need to be one server, right? Potentially, you'll integrate with multiple different services or have a workforce of various examples that you can leverage to interact with all of these different servers that have different capabilities. It's a really interesting release, and it will be fascinating to see over the weeks and months following if this starts to gain adoption. They have a bunch of really great examples, but I just wanted to create a quick video pointing you in the direction and letting you know about this new model context protocol that's out there.

Let me know your thoughts in the comments. Are you going to be using this? How do you find it better than what we had previously? Please share any of your thoughts in the comments below. If you found this video useful, please like, comment, share, and subscribe. Otherwise, until the next one!