What is MCP?

MCP (Model Context Protocol) is an open standard for connecting AI assistants to external systems. Think of it as a universal language that lets AI models talk to APIs, databases, and services.

The Protocol

MCP defines how AI systems can:

1. Discover tools - Ask "what can you do?"
2. Call tools - Request "do this specific thing"
3. Access resources - Read data and context
4. Receive prompts - Get specialized instructions

Core Concepts

Tools

Tools are functions the AI can call. Each tool has:

{
  name: "search_orders",           // Unique identifier
  description: "Search orders...", // Helps AI decide when to use it
  inputSchema: {                   // JSON Schema for parameters
    type: "object",
    properties: {
      query: { type: "string" }
    },
    required: ["query"]
  }
}

The AI reads the description to understand when to use the tool, and the inputSchema to know what arguments to provide.

Resources (Optional)

Resources are read-only data sources:

{
  uri: "config://app-settings",
  name: "Application Settings",
  mimeType: "application/json"
}

Resources are less common than tools. Most integrations only need tools.

Prompts (Optional)

Pre-defined prompt templates for specific tasks:

{
  name: "analyze_order",
  description: "Analyze an order for issues",
  arguments: [
    { name: "orderId", required: true }
  ]
}

Protocol Messages

Tool Discovery

Client                    MCP Server
  │                          │
  │── tools/list ───────────▶│
  │                          │
  │◀── { tools: [...] } ─────│
  │                          │

Tool Execution

Client                    MCP Server
  │                          │
  │── tools/call ───────────▶│
  │   { name: "search",      │
  │     arguments: {...} }   │
  │                          │
  │◀── { content: [...] } ───│
  │                          │

Transport Layers

MCP works over different transports:

Transport	Use Case	How It Works
stdio	Local processes	stdin/stdout pipes
SSE	HTTP servers	Server-Sent Events
WebSocket	Real-time	Bidirectional

Diosc uses SSE because it works well with HTTP infrastructure (load balancers, proxies, etc.).

MCP vs REST APIs

Aspect	REST API	MCP
Client	Any HTTP client	AI assistant
Discovery	OpenAPI/Swagger (optional)	Built-in tools/list
Input validation	Custom	JSON Schema
Purpose	General CRUD	AI-specific actions
Semantics	HTTP verbs	Tool calls

MCP is designed for AI consumption. The descriptions help the AI understand when to use each tool, not just how.

Example: REST to MCP

Original REST API

GET /api/orders?customer=John&status=pending
Authorization: Bearer <token>

MCP Wrapper

// Tool definition
{
  name: "search_orders",
  description: "Find orders by customer name or status. Use this when the user asks about orders, order history, or wants to look up specific orders.",
  inputSchema: {
    type: "object",
    properties: {
      customer: {
        type: "string",
        description: "Customer name to search for"
      },
      status: {
        type: "string",
        enum: ["pending", "shipped", "delivered"],
        description: "Filter by order status"
      }
    }
  }
}

// Tool implementation
async function searchOrders(args, authHeaders) {
  const url = new URL('https://api.example.com/orders');
  if (args.customer) url.searchParams.set('customer', args.customer);
  if (args.status) url.searchParams.set('status', args.status);

  const response = await fetch(url, {
    headers: authHeaders  // Forward user's auth
  });

  return response.json();
}

The key additions:

• Description tells AI when to use this tool
• Parameter descriptions help AI construct valid calls
• Auth forwarding preserves user permissions

Why MCP Matters

For AI Developers

• Standard protocol means one integration pattern
• Tool discovery is automatic
• Clear contract between AI and services

For API Owners

• Expose APIs to AI without rewriting
• Security model unchanged
• Control what AI can access via tools

For Users

• AI can do more things
• Actions are transparent (tool calls are visible)
• Same permissions as direct access

Learn More

• MCP Specification - Official protocol spec
• MCP SDK - Reference implementation
• Diosc Requirements - What Diosc adds to MCP