MCP - What is it?

If you ask an AI Model about something, it looks into what it has already “read” and gives you output. But if it wants something real-time, it should talk to some 3rd party apps or tools and fetch the data. For example, if you ask a model, what is weather like today, or can you look up my outlook calendar and see my availability in next week, or look at this stock price and buy when it reaches at this price,or look for my JIRA tickets that are open, it doesn’t know how to do it, on it’s own.

So it must be integrated with other tools and apps. The way the AI model should be configured and integrated with is different for different apps/tools. It is a lot of work and need lot of customization. There can be 100s of such apps/tools and 100 such customizations.

So here comes MCP. MCP stands for Model Context Protocol. It is a standardized way developed by Anthropic for AI models to talk to different various tools, data sources and applications. Think of MCP like a USB-C port for AI applications. Just as USB-C provides a standardized way to connect your devices to various peripherals and accessories, MCP provides a standardized way to connect AI models to different data sources and tools.

MCP acts as that translator, taking requests from an AI Model and converting them into a protocol/ common language that a specific tool or application can understand. It converts your natural language request, i.e, in plain english, into an API call that the tool can understand.

You do not need 100 diff customizations or code configs for 100 diff tools, you just can leverage MCP and configure all the tools in easy way.

So what happens in background is, when you ask an AI model, how is weather today, it first looks into what tools are available, and which tool can provide this info/help, and then after finding a suitable tool, it translates the request into the tool API language and fetches the data, and converts into user response format that user has asked for. But for that to happen you need an MCP server.

What is an MCP Server?

MCP servers are like bridges between AI models and 3rd party apps that models make use of to serve requests.

They expose tools, resources, prompt templates that clients like AI models can discover and use.

They handle requests for prompts, which are pre-written templates that can help users accomplish specific tasks.

They manage resources (like files, databases, API responses) with URI-based access patterns.

Examples:

GitHub MCP Server: Enables AI to interact with GitHub repositories, allowing for code searching, file updates, and more

Terraform MCP Server: Integrates with Terraform to enable advanced automation and interaction capabilities for Infrastructure as Code (IaC)

Atlassian Remote MCP Server: Allows AI to interact with Atlassian tools like Jira and Confluence

Zerodha Kite MCP Server: Allows AI to interact with Kite API and get real time market info and portfolio insights.