Executive Overview
The Model Context Protocol (MCP) is the universal open standard connecting AI agents to external tools, data sources, and services. Learn how to architect remote MCP servers, advertise discovery cards, and enforce granular authorization.
1. MCP Protocol Fundamentals (Tools, Prompts, Resources)
The Model Context Protocol establishes three standard primitives:
- Tools: Callable functions exposed with JSON Schema contracts that agents can invoke to perform side-effects or query databases.
- Resources: URI-addressable static or dynamic data streams (files, database tables, log feeds) providing contextual knowledge to the model.
- Prompts: Pre-engineered prompt templates with parameter substitution exposed directly by servers.
SEP-1649 MCP Server Card Manifest
{
"$schema": "https://modelcontextprotocol.io/schemas/server-card-v1.json",
"serverInfo": {
"name": "tinycto-tv",
"version": "2.3.0",
"description": "Production MCP server for AEO auditing and AI Agent Readiness scoring"
},
"transport": {
"type": "sse",
"endpoint": "https://tinycto.tv/api/mcp",
"url": "https://tinycto.tv/api/mcp/sse",
"protocols": ["jsonrpc-2.0"]
},
"capabilities": {
"tools": { "listChanged": false }
}
}2. Agent Skills Discovery Index (RFC v0.2.0)
Rather than flooding an agent's context window with hundreds of static JSON schemas, the Agent Skills Discovery RFC allows agents to probe a domain's capabilities on-demand via /.well-known/agent-skills/index.json.
Each skill entry advertises a cryptographic SHA-256 digest, capability description, and execution endpoint URL.