Designing for Context in AI Products: A Guide to MCPs

Let’s unpack what MCPs really are and why they matter in AI product design.

While designing for AI products, understanding the context and presenting information to users in the right context is the key to delivering a good experience. While researching this topic, I found most articles were quite complex and technical to understand. So I decided to write this blog to help demystify the concept so we know what MCPs are and how they matter in AI product design.

Context exists in multiple places. For example, let’s say you design an AI personal agent (for simplicity, you can also think of it as AI assistant). Your agent needs to gather information or context to best assist you — it needs to connect to your calendar, your inbox, even your Notion notes or meeting notes, etc.

All these tools can be scattered in multiple places without having one interface or standardized way for AI applications or agents to access them.And that’s where MCPs come into the picture. So let’s understand what an MCP is.

So whats a MCP?

Now that you know what problem MCP address, let’s understand what an MCP is in the first place. MCP stands for Model Context Protocol. Model refers to LLMs or intelligent agents, context refers to memory and user-specific information, and protocol refers to the rules that define how context is structured and exchanged.

Why rules?

Without rules, your AI product, is like a human who doesn’t know how to behave. Like if a human is in a library, they act differently compared to when they are at a party. Similarly, these protocols act like a rule system that helps AI perform better in the right context.

These rules also help with the security and safety of AI products.

Without MCP, AI products would act like an assistant that gives you a pile of documents when you ask for something specific. MCP servers ensure your assistant gives you the right document for the right meeting. So it’s an AI model, with the right context and rules.

So how is this applied and leveraged in real products?

Figma recently launched a Dev Mode MCP server, which enables AI tools (like Copilot or Cursor running inside VS Code) to access Figma design files, frames, variables, and tokens — giving those tools the right context while coding. The MCP server runs locally and acts as a structured interface between the design file and the AI agent. It’s not a memory store itself, but a gateway to well-structured design data.

So next time you hear someone say "MCP server" — think of it as a gateway to structured data for AI models.

Thank you reading

How can I further help you?

👉 Get your Free 7-Day AI Product Design Roadmap: https://theaidesignacademy.com/roadmap