MCP Explained: A simple guide for product teams
🧠 Knowledge Series 66: Everything you need to know about this new protocol that everyone’s talking about
🔒The Knowledge Series is available for paid subscribers. Get full ongoing access to 60+ explainers and tutorials to grow your technical knowledge at work. New guides added every month.
OpenAI’s CEO Sam Altman has confirmed that it is now supporting the new protocol that everyone’s talking about: MCP. Here’s what he had to say:
Microsoft has also recently confirmed that its new Copilot Agent Studio will support MCP.
But what exactly is MCP and why should product teams care about it?
In this Knowledge Series, we’ll cover the essentials worth knowing from a non-engineer’s perspective. We’ll explore some of the high level technical concepts involved in MCP, why it matters to product teams along with some real world examples from leading companies like Block and Microsoft. Plus, practical ways MCP can be used during the product development process with tools like Figma and Jira.
If you’re currently building out AI assistants or agents in your own product and you want to give them the ability to interact with tools and resources or if you’re just interested in staying up to date with some of the most important emerging technologies in AI then this edition of the Knowledge Series should help.
Coming up:
What exactly is MCP?
Why does it matter to product teams?
An end to end example to learn the core concepts
How companies are using MCP: real world examples of MCP in action and use cases from Block, OpenAI and Microsoft
How you can use MCP during the product development process for testing and integrating Figma and Jira into engineering tools
Tools, terminology and resources to get up to speed
What exactly is MCP? A simple explanation
Let’s start with the basics. The tech industry loves an acronym and MCP is an acronym for Model Context Protocol. It sounds pretty complex but if we break this down into its individual parts it starts to make a little more sense:
M is for Model – typically an LLM like GPT-4, Claude, or Gemini.
C is for Context – the external data, tools, or resources the model needs to complete a task. So, for example, you might want your travel chatbot / agent to be able to access external abilities like searching the web, accessing databases or tools that convert currencies.
P is for Protocol – the rules and message format that define how clients (like chat UIs) and servers (which hold the data or tools) talk to each other.
If you think about it, LLMs are powerful, but they really can’t do much without further context. If I asked an LLM to do this, it wouldn’t be able to do it without more context:
“Check if the data included in my presentation matches the total in the SQL database”An LLM can’t do that on its own. It needs:
Access to your database
Access to your presentation file
MCP helps the model connect to the tools or resources that provide that context that it needs and the protocol makes this easier for engineers.
What are protocols exactly?
Protocols are designed to standardize the rules of data exchange to make it easier to connect different pieces of technology together.
In this case, MCP is an open source protocol developed by Anthropic. If you’ve worked with APIs then you’ll know that many APIs work with REST protocols. Here’s a snapshot of some of the common protocols you may have used in the past:
HTTP - The foundation of data communication on the web, with HTTPS adding encryption for security.
REST - An architectural style that uses standard HTTP methods for communication in APIs.
FTP - For transferring files between systems.
The MCP Protocol gives developers a standardised way for product teams to connect their AI assistants with external things like databases, tools, APIs and any other resource that might be useful. Before MCP, bespoke integrations were required for each tool, but now a standardised protocol (MCP) exists for engineers to quickly integrate with various different tools at once.
We’ll take a look at some real world examples of how companies are using this together later.
The core technical components of an MCP set up include: