MCP in One Sentence
MCP gives AI applications a standard way to connect to tools, files, and services.
AI host -> MCP client -> MCP server -> tools / data
MCP Cheat Sheet
A quick reference for the Model Context Protocol: what it is, how it works, and why it matters for AI-assisted QA workflows.
The Three Roles
These roles appear in nearly every MCP explanation and setup flow.
Host: the AI app Client: protocol handler inside the host Server: exposes tools, resources, prompts
Tool Call Flow
This is the practical sequence behind “AI using tools.”
1. Client asks server what tools exist 2. Model chooses a tool 3. Client sends tool call with arguments 4. Server executes 5. Result is returned to the host
Common MCP Concepts
These are the building blocks students should remember.
Tool: A callable capability like browser_click Resource: Data the server can expose Prompt: A reusable template supplied by the server
Transports
Different environments use different transport styles.
stdio: - local child process - common for desktop tools HTTP/SSE: - remote service - useful for shared or hosted tools
QA Use Cases
MCP is especially powerful when AI needs structured access to testing systems.
- browser automation - code review helpers - Jira integration - file system exploration - CI failure diagnosis - report generation
Tool Schema Mindset
Each tool should be clear, constrained, and easy for the model to call correctly.
Good tool schema: - strong name - clear description - typed arguments - safe behavior - predictable return shape
Security
Tool access should be deliberate, not open-ended.
- use least privilege - review write access carefully - avoid exposing secrets - keep auditability where possible - prefer read-only tools unless writes are required
MCP vs Ad-hoc Integrations
The benefit is standardization, not magic.
Without MCP: - custom integration per tool With MCP: - one protocol - discoverable tools - shared structure - easier multi-tool orchestration