mcp-servers-guide

MCP Servers Comprehensive Guide

MCP (Model Context Protocol) is the open standard for connecting AI agents to tools. This guide covers everything from architecture to deployment.

What Is MCP?

AI Agent (Claude, GPT, etc.)
    │
    ▼
MCP Client (built into host app)
    │
    ▼
MCP Server (your tool/service)
    │
    ├── Tools (functions the agent can call)
    ├── Resources (data the agent can read)
    └── Prompts (templates for the agent)

Core Primitives

Tools

Functions the agent can execute:

server.tool('getTokenPrice', {
  description: 'Get the current price of a cryptocurrency',
  parameters: z.object({
    symbol: z.string().describe('Token symbol (e.g., SPA, ETH)'),
    currency: z.string().default('usd').describe('Target currency')
  }),
  handler: async ({ symbol, currency }) => {
    const price = await fetchPrice(symbol, currency);
    return { content: [{ type: 'text', text: `${symbol}: $${price}` }] };
  }
});

Resources

Data the agent can read:

server.resource('portfolio', {
  description: 'User DeFi portfolio',
  uri: 'portfolio://current',
  handler: async () => {
    const portfolio = await getPortfolio();
    return { content: [{ type: 'text', text: JSON.stringify(portfolio) }] };
  }
});

Prompts

Templates for the agent:

server.prompt('defi-analysis', {
  description: 'Analyze a DeFi protocol',
  arguments: [{ name: 'protocol', description: 'Protocol name' }],
  handler: ({ protocol }) => ({
    messages: [{
      role: 'user',
      content: `Analyze the DeFi protocol "${protocol}" covering TVL, APY, risks, and team.`
    }]
  })
});

Transport Types

Transport	How It Works	Best For
stdio	Command-line process (stdin/stdout)	Local tools, CLI
SSE	Server-Sent Events over HTTP	Remote servers
Streamable HTTP	HTTP with streaming	Web services

stdio Setup

{
  "mcpServers": {
    "my-tool": {
      "command": "npx",
      "args": ["@my/mcp-server"]
    }
  }
}