Getting Started with MCP (Model Context Protocol)
What is the Model Context Protocol?#
The Model Context Protocol (MCP) is a specification originally created by Anthropic that defines how AI agents discover and interact with external tools and services. Think of it as an API catalog specifically designed for AI systems. When an AI agent encounters a task that requires external capabilities — checking a price, placing an order, querying a database — it can read your MCP manifest to understand what tools you offer, what parameters they accept, and how to call them. MCP has been donated to the Agentic AI Foundation (AAIF) under the Linux Foundation, with platinum members including Anthropic, OpenAI, AWS, Google, Microsoft, Bloomberg, and Cloudflare. This governance structure means MCP is becoming the de facto standard for AI-to-service communication.
MCP vs Traditional APIs#
Traditional REST APIs are designed for human developers who read documentation, understand authentication flows, and write integration code. MCP is designed for AI agents that need to dynamically discover and use tools without prior integration. The key differences are: MCP manifests are machine-readable by design, with explicit parameter schemas and natural-language descriptions that AI models can interpret. MCP supports real-time bidirectional communication via Server-Sent Events (SSE), enabling streaming responses and long-running operations. Traditional APIs require custom client code for each integration, while MCP provides a universal interface that any MCP-compatible agent can use immediately.
MCP does not replace your existing APIs. It provides a discovery and interaction layer on top of them, making your existing services accessible to AI agents.
Creating Your MCP Manifest#
The MCP manifest is a JSON file placed at /mcp.json or /.well-known/mcp.json on your domain. It describes your available tools, their parameters, and how to invoke them. Each tool has a name, description (in natural language that AI can understand), input schema (JSON Schema format), and an endpoint URL. Below is an example manifest for a domain intelligence service.
{
"schema_version": "1.0",
"name": "Citability Domain Intelligence",
"description": "AI visibility scoring and domain intelligence API",
"tools": [
{
"name": "scan_domain",
"description": "Scan a domain for AI visibility across 62 protocol endpoints. Returns scores for Discovery, Agentic Commerce, and Citation Readiness.",
"inputSchema": {
"type": "object",
"properties": {
"domain": {
"type": "string",
"description": "The domain to scan (e.g. example.com)"
}
},
"required": ["domain"]
},
"endpoint": "/api/v1/scan"
},
{
"name": "get_domain_authority",
"description": "Get authority scores and crawl metrics for a domain from 607M+ domain dataset.",
"inputSchema": {
"type": "object",
"properties": {
"domain": {
"type": "string",
"description": "The domain to look up"
}
},
"required": ["domain"]
},
"endpoint": "/api/v1/domain"
}
],
"authentication": {
"type": "api_key",
"header": "X-API-Key"
}
}Governance and the AAIF#
In early 2026, MCP was donated to the Agentic AI Foundation (AAIF) under the Linux Foundation. This was a significant milestone because it means MCP is no longer controlled by a single company. The AAIF's platinum members — Anthropic, OpenAI, AWS, Google, Microsoft, Bloomberg, and Cloudflare — collectively guide the protocol's evolution. For businesses implementing MCP, this governance structure provides confidence that the standard will be maintained, improved, and widely adopted. The AAIF also governs AGENTS.md (for AI coding agents) and the Goose framework, creating a coherent ecosystem for agentic AI standards.
Frequently Asked Questions
No. MCP is a discovery layer on top of your existing APIs. Your mcp.json manifest simply describes your existing endpoints in a format that AI agents can understand. Your actual API implementation does not change.
Claude (via Claude Desktop and the API), ChatGPT (with tool use), and many open-source agents support MCP. As the AAIF expands, MCP support is becoming standard across all major AI platforms.
MCP is the leading standard for AI-to-service communication, but it is not the only one. Google's UCP and Stripe/OpenAI's ACP are complementary protocols focused specifically on commerce. Implementing MCP alongside these gives you the broadest agent compatibility.