Model Context Protocol (MCP)
The Problem MCP Solves
Before MCP, every AI app integrated with every tool through bespoke glue code — N×M integrations. MCP is an open standard (originally from Anthropic, now broadly adopted) that gives models a uniform way to discover and call tools, retrieve resources, and use prompts from external servers.
How It Works
An MCP server exposes capabilities — tools (callable functions), resources (data), prompts (templates). An MCP client (the AI host application) connects, negotiates, and calls them through a standardized JSON-RPC protocol. Auth, schemas, and discovery are all built in.
Why It Matters for Security
MCP is becoming the de facto integration fabric for agentic systems. Major AI and security platforms, from Anthropic, OpenAI, Google, Microsoft and others, now expose MCP servers or consume them. That means a platform's openness is increasingly measured by its MCP support — and so is its attack surface.
The Security Angle
Every MCP connection is a potential injection vector. A malicious MCP server can poison context, exfiltrate data, or manipulate tool calls. Questions to ask of any MCP deployment: How do you authorize MCP servers? Do you scan tool definitions for injection? Where does the audit log live for agent-to-tool calls?
Prefer slides, quizzes, and saved progress? Read this lesson in thelibrary — free, no sign-up.