Why product people should know about MCP (and what they should know about it)
I recently did an agentic AI course in which I learned how to create agents and connect them in a workflow. The need for solid data was one of the things that this course reinforced, with a strong focus on knowledge bases and Retrieval-Augmented Generation (RAG).
I’ve since then learned about the Model Context Protocol (MCP) and how this new API protocol will make it easier for AI systems to access the data that they need. I’ll share my early learnings about MCP and why I believe that this isn’t something that product people can ignore.
MCP — The ‘what’
MCP is an open protocol that enables developers to build two-way connections between AI Models (LLMs) and the Context they need to accomplish tasks. For example:
- AI assistant providing healthcare diagnosis — Context would require relevant patient medical history, lab results, etc.
- AI assistant analysing legal documents — Context would require specific case law references, regulatory frameworks, etc.
- AI assistant optimising financial portfolios — Context would need market data, relevant investment goals, risk profile, etc.
The most common Context Types are “tools” and “resources”. Tools are the functions that the LLM can use to perform specific tasks, for example:
- Database query tools
- Specialised search capabilities
- Tools for complex calculations
Resources are the information attachments that are provided to the LLM, for example:
- Company knowledge bases
- User or transaction specific data
- Multi-modal content
From an architecture perspective, there are MCP Servers that handle requests from MCP Clients. Clients such as Windsurf and Cursor can fire reflection requests to ask the server for information about the tools and resources it has access to.
Let’s take a simple real-life example to better understand how MCP would work. Let’s say I’ve got a chronic health condition and I’m using an AI health assistant to help me with my specific situation.
I ask the assistant: “I’ve been feeling more tired than usual this week. Should I adjust my medication or routine?”
My health assistant (MCP Client) communicates with multiple MCP Servers including:
- Electronic Health Record (EHR) systems
- Medication information databases
- Symptom analysis engines
When I make my request, the system follows a structured process:
- The MCP host identifies which health contexts are relevant
- The AI pulls my medication history and potential interactions from my EHR
- It analyses sleep patterns from my sleep tracking app
- All this information is then synthesised into a personalised health assessment
For the AI to provide me with an accurate health assessment, it needs to access the right context, i.e. the relevant tools and resources. Instead of having to integrate each health data source separately, the health assistant uses MCP to seamlessly access diverse health information through a standardised protocol.
MCP — The ‘why’
Though in its early stages, MCP deserves product teams’ attention for two compelling reasons:
- Enhanced AI capabilities — MCP extends LLMs beyond their training data through structured access to external tools and real-time information, enabling more complex tasks and contextually relevant responses.
- Universal integration standard — MCP eliminates bespoke connections between LLMs and applications by establishing a common communication protocol.
In short, MCPs will make LLMs much more capable, enabling more powerful and context-rich applications and end-user experiences.
Main learning point: As product people we need to keep following how the MCP standard evolves, learn how platforms use this new standard and select the right MCP server(s) to support a specific use case.
Related links for further learning:
- https://www.linkedin.com/pulse/model-context-protocol-mcp-ai-software-development-piyush-ranjan-dxj0e/
- https://modelcontextprotocol.io/introduction
- https://docs.cursor.com/context/model-context-protocol
- https://blog.dailydoseofds.com/p/visual-guide-to-model-context-protocol
- Building Agents with Model Context Protocol — Workshop with Mahesh Murag of Anthropic
- https://norahsakal.com/blog/mcp-vs-api-model-context-protocol-explained/
- https://huggingface.co/blog/Kseniase/mcp
- https://www.thecloudgirl.dev/blog/what-is-mcp
