Model Context Protocol (MCP) Documentation
by Model Context Protocol
The primary specification for MCP — and the place the 2026 stateless redesign is actually documented.
Overview
The official documentation for the Model Context Protocol, the open standard for connecting AI applications to external systems — described on its own front page as a USB-C port for AI applications. The site splits into a getting-started track, a Learn section covering architecture and server and client concepts, Develop guides for building servers and clients, and the versioned specification itself. The architecture page is the one worth reading closely: MCP is a client-server design in which a host application creates one MCP client per server, communicating over either the stdio transport for local processes or Streamable HTTP for remote ones, with OAuth recommended for obtaining authentication tokens. The data layer is JSON-RPC 2.0. Servers expose three primitives — tools (executable functions), resources (contextual data sources) and prompts (reusable interaction templates) — each with list-style discovery methods and, for tools, a tools/call execution method. On the client side, elicitation lets a server request additional input or confirmation from the user. Notably, the 2026-07-28 revision makes the protocol stateless: every request carries protocol version and capabilities in a _meta field, a mandatory server/discover request replaces connection-time handshaking, change notifications move to an opt-in subscriptions/listen stream, and both sampling and logging are deprecated. Official SDKs are published in tiers: TypeScript, Python, C# and Go at Tier 1; Java, Rust and Ruby at Tier 2; Swift, PHP and Kotlin at Tier 3.
At a Glance
- Topic
- MCP
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~2-3 hours to work through the concepts and build a first server
- Provider
- Model Context Protocol
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Distinguish MCP hosts, clients and servers and how each connection is established
- ✓Choose between the stdio and Streamable HTTP transports for a given deployment
- ✓Expose tools, resources and prompts as server primitives with JSON Schema inputs
- ✓Handle the stateless request model and the mandatory server/discover exchange
- ✓Use elicitation to request additional input or confirmation from the user
- ✓Subscribe to change notifications through the opt-in subscriptions/listen stream
- ✓Pick an official SDK with full knowledge of its support tier and feature completeness
Highlights
- •The primary specification — every other MCP tutorial and course is downstream of this site
- •Full JSON-RPC request and response examples for discovery, listing, tool calls and notifications
- •Versioned documentation, so you can read the exact revision your SDK actually implements
- •Explicitly marks sampling and logging as deprecated in 2026-07-28, which is a live migration trap
- •Anthropic donated the protocol to the Agentic AI Foundation in December 2025, so it is no longer single-vendor governed
Who It’s For
Best For
- ✓Backend engineers building an MCP server over internal data or tools
- ✓Application developers adding MCP client support to an AI product
- ✓Architects evaluating MCP against bespoke tool-calling integrations
Prerequisites
- •Comfortable with JSON-RPC or a similar request/response RPC model
- •Working knowledge of one SDK language: TypeScript, Python, C#, Go or Java
- •Understanding of how LLM tool calling works
FAQ
What is Model Context Protocol (MCP) Documentation?
The official documentation for the Model Context Protocol, the open standard that lets AI applications connect to external tools and data. It covers the client-server architecture, both transports, the JSON-RPC data layer and every server and client primitive, alongside build guides and tiered official SDKs for ten languages.
Is Model Context Protocol (MCP) Documentation free?
Model Context Protocol (MCP) Documentation is free to access.
What level is Model Context Protocol (MCP) Documentation for?
Model Context Protocol (MCP) Documentation is aimed at a intermediate audience. Recommended background: Comfortable with JSON-RPC or a similar request/response RPC model, Working knowledge of one SDK language: TypeScript, Python, C#, Go or Java, Understanding of how LLM tool calling works.
How long does Model Context Protocol (MCP) Documentation take?
Expect roughly ~2-3 hours to work through the concepts and build a first server. Most learners work through it at their own pace.
What will I learn from Model Context Protocol (MCP) Documentation?
You'll learn: Distinguish MCP hosts, clients and servers and how each connection is established; Choose between the stdio and Streamable HTTP transports for a given deployment; Expose tools, resources and prompts as server primitives with JSON Schema inputs; Handle the stateless request model and the mandatory server/discover exchange; Use elicitation to request additional input or confirmation from the user; Subscribe to change notifications through the opt-in subscriptions/listen stream; Pick an official SDK with full knowledge of its support tier and feature completeness.
Topics
Sources
This page was written from 3 sources, 2 on domains other than modelcontextprotocol.io.