Model Context Protocol (MCP)

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open specification, introduced by Anthropic, for connecting AI agents and LLM applications to the outside world — tools they can call, data they can read, and context they can use. It answers a narrow but common question: how should an agent talk to a tool?

Rather than writing a bespoke adapter for every integration, a developer runs (or connects to) an MCP server that exposes capabilities in a standard shape. Any MCP client — an IDE, a chat app, an agent runtime — can then use those capabilities without custom code.

The client–server model

MCP uses a client–server architecture communicating over JSON-RPC 2.0. The host application spins up MCP clients, each of which connects to an MCP server.

• Host — the AI application (e.g. an IDE assistant or chat client) the user interacts with.
• Client — a connector inside the host that maintains a session with one server.
• Server — a program exposing tools, resources and prompts for a specific system (a database, a file store, an API).

The three primitives: tools, resources, prompts

An MCP server offers up to three kinds of capability. Keeping them distinct is what makes the protocol predictable.

• Tools — actions the model can invoke (e.g. query_database, create_ticket). These map naturally onto model function calling.
• Resources — read-only data the client can load into context (files, records, documents).
• Prompts — reusable, parameterised prompt templates a server can offer to guide common tasks.

// Client asks a server what it can do (JSON-RPC request)
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list"
}

// Server responds with its tool catalogue
{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "tools": [
      {
        "name": "search_docs",
        "description": "Full-text search across the knowledge base",
        "inputSchema": { "type": "object", "properties": { "query": { "type": "string" } } }
      }
    ]
  }
}

When to use MCP

Reach for MCP when the core need is connecting an agent to tools, data or context — especially if you want that connection to be reusable across clients or shareable with others.

• Giving an agent access to internal systems (databases, ticketing, file stores) through a consistent interface.
• Building a tool once and letting many different AI apps use it.
• Standardising tool access across a team so integrations aren't rewritten per project.

MCP compared to A2A and Agent Protocol

MCP sits at the tool/context layer. A2A sits at the agent-to-agent layer. The Agent Protocol REST spec sits at the client-to-agent layer. They compose. See the full comparison for details.