Model Context Protocol Explained (For Developers)

Note: This page explains the low-level details of the MCP protocol.
For the practical OpenCode setup guide, servers list, and usage, please visit the authoritative hub:
→ OpenCode MCP Guide

What describes MCP technically?

The Model Context Protocol (MCP) is an open standard that enables AI models to interact with external data and tools through a standardized interface. It is built on top of JSON-RPC 2.0.

Unlike proprietary plugin systems, MCP decouples the Client (the editor/IDE) from the Server (the tool provider).

Core Architecture

The protocol defines three primary primitives:

Prompts: Pre-defined templates that servers can expose to clients.
Resources: Data sources that can be read by clients (like files, logs, or database rows).
Tools: Executable functions that can be invoked by the model (like "execute_command" or "query_db").

Message Flow (JSON-RPC)

Communication happens via standard JSON-RPC messages.

Request (Client -> Server):

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "read_file",
    "arguments": {
      "path": "/home/user/project/README.md"
    }
  }
}

Response (Server -> Client):

{
  "jsonrpc": "2.0",
  "id": 1,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "# Project Documentation..."
      }
    ]
  }
}

Protocol Capability Negotiation

When a client connects to an MCP server, they perform a handshake to negotiate capabilities.

Client Capabilities: Does the client support sampling? Does it support notifications?
Server Capabilities: Does the server expose resources? Prompts? Tools?

This negotiation ensures backward and forward compatibility as the protocol evolves.

Why this abstraction matters

For developers building tools, MCP eliminates the need to write specific adapters for every IDE (VS Code, JetBrains, OpenCode). You write one MCP server, and it works with any MCP-compliant client.

Introduction

Oh My OpenCode

Getting Started

Guides & Integrations