MCP Without the Hype
MCP gives AI applications a common way to discover and use capabilities exposed by other systems. It does not decide whether an operation is safe, who is allowed to perform it, what data may cross the boundary or who owns the server after the demo. This six-part series explains MCP in plain language, separates it from APIs and function calling, treats it as a context and authority boundary, shows how to design a server as a product, provides a reusable security review and weighs when an enterprise gateway helps. One fictional South African insurer carries the examples, with protocol details checked against the current specification.
6 articles
- Part 1MCP in plain language: it connects capability, it does not establish trustThe Model Context Protocol is often sold as the universal adapter that makes AI agents enterprise-ready. Part 1 explains what MCP actually standardises, what it leaves to you, and why a connector that works is evidence of connection, not of safety.8 min read
- Part 2API, function calling and MCP are different layersTeams argue about whether MCP replaces their APIs, or whether function calling makes MCP unnecessary. Part 2 separates the three: what each layer does, what each one leaves out, and why the authorisation work underneath all of them never goes away.7 min read
- Part 3MCP is a context boundary, not a tool catalogueThe first question teams ask about MCP is how many tools the agent can see. The better question is what context and authority cross the boundary in each direction. Part 3 covers tools, resources and prompts as different trust questions, and why a tool description is part of the attack surface.7 min read
- Part 4Design an MCP server as a product, not a wrapperThe fastest way to build an MCP server is to wrap an existing API, endpoint by endpoint. It is also the fastest way to create an unowned, over-powered integration nobody can support. Part 4 covers naming, schemas, errors, idempotency, limits, versioning and the owner every server needs after the demo.8 min read
- Part 5The MCP security review: nine questions before a server goes liveAn MCP server can follow the specification to the letter and still let one claims handler read another's claims. Part 5 is a reusable security review: nine questions, the failure each one catches, and how controls should scale with the consequence of each tool.8 min read
- Part 6The enterprise MCP gateway: control point or new bottleneck?Once several AI applications reach several MCP servers, a central gateway for discovery, credentials, quotas, audit and a kill switch starts to look essential. It can also become the queue every new tool waits in. Part 6 covers when a gateway earns its place, what it should and should not do, and how to keep it from becoming the bottleneck.8 min read
Working through any of this in your own organisation?
These articles come out of delivery work rather than theory. If the problems in this series look familiar, a short scoping conversation is usually the quickest way to work out what is worth doing first.
Talk to CloudNala