MCP Introduction

MCP stands for Model Context Protocol. It solves a practical agent problem: when an agent needs tools, files, databases, search results, or business systems, context should not be assembled through one-off prompt text. Capabilities, data, and safety boundaries should be exposed through a clear protocol.

What you will learn

• What client, server, tool, resource, and prompt mean in MCP.
• How to design an external capability as a discoverable, callable, testable interface.
• How to validate MCP tool responses and errors with a Harness.
• How to define boundaries for permissions, secrets, and dangerous operations.

Minimal mental model

Role	Purpose
Client	The agent or host app that discovers and calls server capabilities
Server	The capability provider exposing tools, resources, and prompts
Tool	A callable action with computation or side effects
Resource	Readable data such as files, records, or document fragments
Prompt	A reusable task instruction or workflow template

Design steps

1. Define the task boundary.
2. Define inputs and outputs.
3. Decide permissions.
4. Write Harness scenarios.
5. Document usage, testing, and rollback.

Practice

Design a search_docs MCP tool:

• Input: query and optional limit.
• Output: matching title, summary, URL, and confidence.
• Acceptance: empty results include a next step; empty query returns a clear error.

Next: Core Concepts.