MCP — The USB-C Standard for AI Tools

When I first built AI agents, every tool integration was custom code I wrote myself.

// I wrote this for every single integration
async function executeTool(name, input) {
  if (name === "check_vpn_status") {
    return await fetch("https://vpn-api.acme.com/...");
  }
  if (name === "check_jira_access") {
    return await fetch("https://jira.acme.com/...");
  }
  // repeat for every tool...
}

Every app needed its own version. Every team duplicated the work.

MCP solves this.

---

What MCP Is

MCP = Model Context Protocol

Created by Anthropic in 2024. Now adopted by OpenAI, Google, Microsoft, and Cursor.

It’s a standard protocol for connecting LLMs to tools — like USB-C, but for AI.

Before MCP:
  Claude → custom code → GitHub
  Claude → custom code → Slack
  Claude → custom code → Postgres
  (every connection different, repeated everywhere)

After MCP:
  Claude → MCP → GitHub MCP Server
  Claude → MCP → Slack MCP Server
  Claude → MCP → Postgres MCP Server
  (one standard, plug and play)

---

What MCP Servers Expose

MCP servers expose three things:

Tools — actions the LLM can call

create_github_issue, send_slack_message, query_database

Resources — data the LLM can read

files, docs, database rows, emails

Prompts — reusable prompt templates

pre-built instructions for common tasks

---

Manual Tools vs MCP

// Before MCP — you write everything
const tools = [{
  name: "create_github_issue",
  description: "...",
  input_schema: { ... }
}];

async function executeTool(name, input) {
  if (name === "create_github_issue") {
    await fetch("https://api.github.com/...", { ... });
  }
}

// After MCP — just connect to the server
mcp_servers: [{
  type: "url",
  url:  "https://github-mcp-server.com/sse",
  name: "github"
}]
// Tools available instantly — no executeTool() needed

---

Building Your Own MCP Server

import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { z } from "zod";

const server = new McpServer({ name: "it-support", version: "1.0.0" });

server.tool(
  "check_vpn_status",
  "Check if employee VPN is active or expired",
  { employee_id: z.string() },
  async ({ employee_id }) => {
    const result = await checkVPN(employee_id);
    return { content: [{ type: "text", text: JSON.stringify(result) }] };
  }
);

server.listen();

Now any MCP-compatible AI can use your VPN tool. Write once. Use everywhere.

---

When to Use MCP vs Manual Tools

	Manual Tools	MCP Server
Learning / prototyping	✅	Overkill
One app uses tools	✅	Overkill
Multiple apps need same tools	❌ Duplicate code	✅
Team sharing integrations	❌	✅
Claude Desktop / Cursor access	❌	✅

I built manual tools first to understand the internals. Then MCP made complete sense — it’s the same concept, standardized.

---

Popular MCP Servers Available Now

Official (Anthropic): Filesystem, Postgres, GitHub, Slack, Google Drive
Community: Jira, Notion, Linear, Stripe, Cloudflare

Claude Desktop, Cursor, and Zed Editor all support MCP. Your MCP server works with all of them out of the box.

---

Key Takeaway

MCP = standard protocol for connecting LLMs to the world

Manual tools taught me WHY tool calling works. MCP is the industry’s HOW at scale.

Think of it like REST APIs — you understood HTTP first, then frameworks made sense. Same pattern here.

Next: embeddings — how text becomes numbers that capture meaning.