David Baddiel Tries to Understand is a BBC Radio 4 series. He investigates a topic suggested by someone on X then plays back his understanding to them. I am curious about an evolving standard called MCP (Model Context Protocol) which could radically simplify the way AI tools are built and used. Hence, my rhetorical question is: “What is MCP, how does it work and how important is it?”. Here’s my Baddiel style response.
What is MCP?
The Model Context Protocol is an open standard that enables developers to build secure, two-way connections between their data sources and AI-powered tools. - Anthropic
Think of MCP as the USB-C of AI. Instead of needing a separate charger for every device, we have one universal standard. MCP works the same way: one protocol lets AI connect to any data source, whether that’s a local file system, a PostgreSQL database or GitHub.
MCP comes in two parts:
The specification: rules for how communication should work.
Implementations: actual software libraries and servers that follow those rules.
Before MCP, connecting 10 AI apps to 20 data sources meant writing 200 bespoke connectors. With MCP, each app and each data source implements MCP once; everything talks to everything. Multiplication becomes addition.
That’s why people are getting excited about it.
MCP building blocks
It provides a universal, open standard for connecting AI systems with data sources. - Anthropic
MCP’s architecture has three main characters:
Host Application: This is the app users interact with: ChatGPT in your browser, Claude Desktop or a custom enterprise tool. The host orchestrates the dance: receiving your question, figuring out what tools are needed and presenting the final answer.
MCP Client: The translator inside the host. If the host needs a database query, it spins up a client to talk to the right server. Each client uses MCP to interface outwards and converts responses into the host’s native format.
MCP Server: The bridge to the real world system. A GitHub server knows how to talk to GitHub’s API. A PostgreSQL server knows SQL. Servers can be local (on our laptop) or remote (in the cloud). Developers, companies and open source contributors can all build them.
How does MCP work?
OpenAI’s support of Anthropic’s Model Context Protocol (MCP) may be the start of easier interoperability among AI agents. - Constellation Research
Let’s trace an example request:
We type “What’s our top-selling product?” into our AI app.
The AI recognises it needs fresh sales data.
The host activates an MCP client for our company’s database.
The client sends a neatly formatted JSON-RPC message: “Get top-selling product”.
The server translates this into SQL, queries the database and retrieves the answer.
Results flow back through MCP and the AI replies: “Product A with £487,000 sales last month.”
Each part does its job. The AI understands language. The client handles MCP. The server deals with the database. None has to know how the others work.
Why is MCP important?
MCP is a good protocol and it’s rapidly becoming an open standard for the AI agentic era. - Demis Hassabis
The internet only became the internet because we agreed on TCP/IP. Web browsers and websites only flourished once we all spoke HTTP. MCP is aiming for the same role in the AI era. It is:
Open source: no company owns it.
Simple: JSON messages we can read by eye.
Scalable: every new server expands what all AIs can do.
Instead of static, frozen-in-time models, MCP turns them into connected assistants that can interact with the world.
If you’re a developer, MCP is young enough that your contribution could shape the standard. If you’re a business, MCP is the thing that might let AI talk fluently to your data without endless bespoke integrations.
Other resources
MCP Explained talk by Shaw Talebi
Blockchain: Fad or Future? post by Phil Martin
How to Design Effective AI Agents post by Phil Martin
MCP is less about making models smarter, more about making them useful. It doesn’t upgrade the brain; it wires it into the world.
Have fun.
Phil…
Phil,
Given the current AI language gap between business executives and technical ranks, this is very useful and simple explanation that everyone can undstand and agree to. Well said.
Pervez