Why MCP?
MCP makes AI development less fragmented by standardizing connections between AI applications and external data sources.
What is MCP?
Some people like to say that models are only as good as the context provided to them. You can have an incredibly intelligent model, but if it doesn't have the ability to connect to the outside world and pull in the necessary data and context, it's not going to be as useful as it can be.
The Model Context Protocol is an open-source protocol that standardizes how your large language model connects and works with your tools and data sources. The idea here is to avoid reinventing the wheel around how we do things like tool use, and instead standardize the way that our AI applications connect with data sources.
In the same way that REST standardizes how web applications communicate with back ends and other systems, the Model Context Protocol wants to achieve the same for AI applications.
The Benefits of Standardization
Let's imagine you're using Claude Desktop, and you ask it a question about retrieving some issues from a GitHub repository. Immediately through natural language you're able to talk to Github. This is the power of MCP. You have connected to an MCP server that's providing the necessary data from GitHub.
You don't need an MCP server to connect Claude to Github. You could also build your own custom app or agent that connects the two. The reality is that, at the moment, everything you can do with MCP can just as easily be done without MCP. The point is building your own custom app every time you want to connect a few different data sources is unsustainable.
Lets say we want to connect our Claude desktop to Asana, through to another MCP server. Asana is a popular project management tool, so now we use natural language to look up issues in GitHub and then assign tickets for particular issues to people in Asana.
As the human in the loop, you can coordinate and verify the actions taken between these two applications with very little code. You're now using natural language to communicating with external data sources with ease.
If we wanted to add a third application to the mix. Without a standardized protocol like MCP, you'd have to figure out where to store this third tool. Where do you store custom schemas related to the new tool? What about the data access layer and authentication logic? Now imagine adding a fourth tool, repeating the whole process over and over again. Many different AI models, each talking to different data sources, each one is connected in a different way.
With MCP, we shift the burden of responsibility and we separate our concerns in a clean, maintainable way. Building and use MCP compatible applications lets us connect them to many different servers for any kind of data access we need. You could have servers for data stores, for customer relationship management tools like HubSpot or Salesforce, or even servers for things like version control.
With MCP, we don't have to learn how to use all these app and processes like Salesforce and Hubspot and versions control. We can use these apps though natural language and orchestrate tasks between them without having to write all the logic ourselves.
Growing Ecosystem
With MCP, there are lots of wins for different audiences. For application developers, connecting to an MCP server involves very little work.
For API developers, building the MCP server once means it can be adopted everywhere.
For users of AI applications, you simply have the data you need from any application you want brought into ChatGPT or Claude. For enterprises and large organizations, the benefits come from separating out concerns and building standalone integrations that different teams can use.
The MCP ecosystem is growing fast. Development is coming from large companies and also startups at the frontier. MCP servers are a lot like working with APIs, they can be private or open source, anyone can build their own or you can use community adopted ones.