MCP Explained for SRE: Model Context Protocol Without the Hype

The simplest mental model

Model Context Protocol (MCP) is a protocol for connecting AI applications to external capabilities and context in a standardized way. Think of it as an interface boundary, not as the intelligence itself.

AI application / host
        ↓
     MCP client
        ↓
     MCP server
      ↙     ↘
   tools   resources

The pieces

• MCP Host: the application environment using MCP.
• MCP Client: the component that connects to a server.
• MCP Server: exposes capabilities through the protocol.
• MCP Tool: an action the model/application can invoke.
• MCP Resource: context or data the application can access.

Why SRE and platform teams should care

MCP can turn one AI application into a dependency graph of remote or local services. That means familiar operational concerns return: discovery, connection failures, timeouts, authentication, rate limits, version compatibility, observability and blast radius.

Security does not disappear behind a protocol

Use Least Privilege for tool permissions and understand MCP Tool Poisoning. Tool descriptions, untrusted resources and external content can influence model behavior. Sensitive actions should have strong authorization and explicit policy boundaries.

Operational checklist

• Know which servers and tools are allowed in production.
• Trace tool name, latency, result status and authorization decision.
• Set timeouts and retry rules per tool.
• Separate read-only context access from state-changing actions.
• Treat server configuration and tool descriptions as security-sensitive supply-chain inputs.