Model Context Protocol (MCP)

The Model Context Protocol (MCP) is an open, vendor-neutral standard introduced by Anthropic in late 2024 to solve the M×N integration challenge between generative AI applications (Claude Desktop, AI IDEs, agentic frameworks) and external data sources (PostgreSQL, GitHub, Slack, local file systems, development tools). Modeled after the transformative Language Server Protocol (LSP) in software development, MCP replaces fragmented, proprietary plugin APIs with a unified, bidirectional protocol communicating via JSON-RPC 2.0.

Validate, test, and debug MCP server manifests with our interactive MCP Schema Validator.

---

Technical Specifications at a Glance

Specification	Details
Originator & Year	Anthropic (November 2024, Open Source / MIT)
Transport Protocols	stdio (Standard Input/Output for local processes) & SSE (Server-Sent Events over HTTP)
Message Protocol	JSON-RPC 2.0
Core Primitives	Prompts, Resources (Data/Content), Tools (Executable Functions)
Supported Clients	Claude Desktop, Antigravity IDE, Cursor, Continue.dev, Zed
Official SDKs	TypeScript (@modelcontextprotocol/sdk), Python (mcp), Kotlin

---

The Architecture of MCP: Clients, Servers, & Hosts

┌────────────────────────────────────────────────────────┐
│                   MCP Host (Application)               │
│               (e.g., Claude Desktop / IDE)             │
│                                                        │
│  ┌──────────────────────────────────────────────────┐  │
│  │                   MCP Client                     │  │
│  └───────┬──────────────────────────┬───────────────┘  │
└──────────┼──────────────────────────┼──────────────────┘
           │ stdio                    │ HTTP / SSE
           ▼                          ▼
┌─────────────────────────┐  ┌─────────────────────────┐
│     Local MCP Server    │  │    Remote MCP Server    │
│  (e.g., Filesystem/Git) │  │  (e.g., Postgres / API) │
└─────────────────────────┘  └─────────────────────────┘

The 3 Core Primitives of MCP

1. Resources: Passive, read-only data sources (such as source code files, database schemas, or application logs) that an AI model can read into its context window.
2. Prompts: Pre-configured, parameterized prompt templates and workflows exposed by the server to guide user interactions.
3. Tools: Executable functions and actions (such as execute_query, git_commit, send_slack_message) that the AI model can invoke with structured JSON Schema parameters after receiving explicit user permission.

---

Anatomy of an MCP Tool Definition (JSON-RPC)

An MCP server exposes available tools by responding to the tools/list RPC endpoint:

{
  "name": "query_database",
  "description": "Executes a parameterized read-only SQL query against the database.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "query": {
        "type": "string",
        "description": "Valid SQL SELECT statement"
      },
      "maxRows": {
        "type": "integer",
        "default": 100
      }
    },
    "required": ["query"]
  }
}

When an agent model chooses to call this tool, the client sends a tools/call request:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "query_database",
    "arguments": {
      "query": "SELECT id, email, created_at FROM users LIMIT 5;",
      "maxRows": 5
    }
  }
}

---

Code Example: Creating an MCP Server in TypeScript

import { Server } from '@modelcontextprotocol/sdk/server/index.js';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
import { ListToolsRequestSchema, CallToolRequestSchema } from '@modelcontextprotocol/sdk/types.js';

const server = new Server(
  { name: 'devflow-mcp', version: '1.0.0' },
  { capabilities: { tools: {} } }
);

// Register available tools
server.setRequestHandler(ListToolsRequestSchema, async () => ({
  tools: [
    {
      name: 'format_json',
      description: 'Format, validate, and beautify a JSON string.',
      inputSchema: {
        type: 'object',
        properties: {
          rawJson: { type: 'string' },
          indent: { type: 'number', default: 2 }
        },
        required: ['rawJson']
      }
    }
  ]
}));

// Connect via standard I/O transport
const transport = new StdioServerTransport();
await server.connect(transport);

---

Frequently Asked Questions

Why is MCP being called the "LSP for AI"?

Just as Microsoft's Language Server Protocol (LSP) allowed any code editor (VS Code, Sublime, Neovim) to support autocomplete and diagnostics for any language without writing $N \times M$ plugins, MCP allows any AI client to connect to any data source or tool ecosystem through one shared, universal standard.

How does MCP handle security and user approval?

MCP enforces human-in-the-loop security boundaries. The host client prompts the user for explicit confirmation before executing sensitive tool calls (like file modifications or shell commands), preventing autonomous rogue agent actions.

How do I validate if my MCP server configuration is correct?

Test your server schemas, method signatures, and tool argument validation rules with our MCP Schema Validator.