Most technology teams understand APIs (Application Programming Interfaces).
Humans, applications and services use APIs every day to exchange information and automate business processes. But what happens when the primary consumer is no longer a human or application developer, but an AI agent?

That’s where the Model Context Protocol (MCP) comes in. MCP is an emerging open protocol designed to allow AI assistants and agents to discover and interact with tools, data sources and services in a consistent way. Rather than hard-coding every integration, MCP provides a standard way for agents to understand what capabilities are available and how to use them.

Think of it as:
– APIs made easier for AI
– A standard “plug-and-play” approach for agents
– A common language between AI models and enterprise systems
– As organisations experiment with AI agents, MCP is rapidly becoming a key part of the conversation.

However, one of the most important lessons is that MCP is not just an integration problem. It is a security and governance challenge as well.
The image below, inspired by an article from Simon Painter, perfectly captures why.

– A browser displays information.
– An AI agent can act on information.

That difference changes everything. An AI agent connected through an MCP server may be able to:
– Read data
– Query systems
– Trigger workflows
– Create or modify information
– Act on behalf of a user

The same untrusted content that a browser merely renders could potentially influence an agent’s actions. That means the controls we’ve learned from web, API and cloud security remain critically important:

– Identity and access management
– Least-privilege permissions
– API gateways and policy enforcement
– Network segmentation
– Audit logging and monitoring
– Supplier and third-party governance
– Data protection controls

My biggest takeaway from the latest work we are doing on MCP is this:
MCP doesn’t replace good architecture. It makes good architecture even more important. The organisations that succeed with AI agents will not be those that deploy the most MCP servers. They will be the ones that build them on strong foundations of security, governance and operational control. I’m interested to hear how others are approaching MCP and agent governance in their organisations.

https://www.simonpainter.com/mcp-security-controls/