MCP — The Standard That Changes How AI Works

What is MCP?

Model Context Protocol (MCP) is an open standard for communication between AI applications and external data sources and tools. Created by Anthropic in November 2024, in December 2025 it was transferred to the Agentic AI Foundation (Linux Foundation).

Analogy: MCP is to AI agents what USB-C is to devices — one standard that connects everything. Instead of building a separate connector for every tool (N×M problem), you build one MCP server and every agent understands it.

Why does MCP matter for QA?

Before MCP, every AI tool had its own way of connecting to external systems. Claude reads Jira? Custom integration. ChatGPT searches Confluence? Another one. Copilot reads logs from Sentry? Yet another.

MCP solves this: you build one MCP server and EVERY AI agent that supports MCP can automatically use it.

Architecture

Your AI application (Host)
    └── MCP Client
         ├── MCP Server: Jira
         ├── MCP Server: GitHub
         ├── MCP Server: Slack
         └── MCP Server: Your own
              └── External systems and data

Three MCP primitives

Primitive	What it does	QA example
Tools	AI performs actions	Create a Jira ticket, run a test
Resources	AI reads data (read-only)	Read a test plan from Confluence, test results
Prompts	Reusable templates	”Generate test cases for this user story”

MCP vs RAG

Feature	MCP	RAG
Direction	Bidirectional — reads AND writes	Unidirectional — reads only
State	Stateful	Stateless
Actions	Modifies data (creates tickets, merges PRs)	Only generates text
Use case	Orchestration of multiple systems	Enriching model knowledge

The best systems use both: MCP for orchestration + RAG for knowledge.

---

In the next post: MCP in Practice — who supports it, QA-relevant servers, a scenario, and security.