MCP (Model Context Protocol)
Standard protocol connecting AI to external tools
MCP (Model Context Protocol) is an open protocol standardizing connections between AI models and external tools/data sources. Previously, each AI and each tool required individual integration (NรM problem), but MCP solves this with a single standard. It has a 3-layer architecture: Host (AI service) โ Client (connection manager) โ Server (tool provider), where Servers provide Tools (executable functions), Resources (reference data), and Prompts (guidelines). It supports two transports: stdio (local) and Streamable HTTP (remote), with 4,000+ MCP servers registered on Smithery.
Key Concepts
Develop MCP Server: wrap external systems (DB, API, etc.) as MCP Tools/Resources
MCP Client connects to Server (stdio or Streamable HTTP)
Server communicates available Tools, Resources, Prompts list to Client
AI model analyzes user request and selects needed MCP Tool
Client executes Server's Tool โ returns results to AI model
AI model utilizes tool execution results to generate final response
Pros
- ✓ Solves NรM integration problem (standardization)
- ✓ Tool developers and AI developers can develop independently
- ✓ Open source + diverse SDKs
- ✓ SaaS deployment possible via Streamable HTTP
Cons
- ✗ Early-stage ecosystem
- ✗ Server security management required
- ✗ Varying MCP support levels across AI models
- ✗ Network dependent (tools unavailable when Server is down)