# MCP 101 Canonical: https://mudpie.ai/101/mcp/ Updated: 2026-09-20 Curator: Ali Abouelatta (https://mudpie.ai/authors/ali-abouelatta/) The protocol's original motivation, current architecture, secure transport and the maintainer decisions that turned a connector format into shared infrastructure. **TL;DR:** MCP standardizes how AI hosts connect to tools, resources and prompts. Read the architecture and transport specs first; then study authorization, client behavior and governance before copying a demo server into production. ## Understand the protocol These sources establish MCP's original problem and its current host-client-server model before implementation details obscure the point. ### [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol) Authority: [Alex Albert](https://mudpie.ai/authors/alex-albert/) **TL;DR:** MCP was introduced as an open standard that replaces one-off data integrations with a common connection between AI applications and external systems. **What you'll read:** The launch thesis, early architecture, initial SDKs and the ecosystem problem Anthropic intended the protocol to solve. ### [MCP architecture](https://modelcontextprotocol.io/specification/2026-07-28/architecture) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** A host owns user experience and security boundaries, clients maintain one server connection each, and servers expose focused context and capabilities. **What you'll read:** The authoritative host-client-server roles, isolation boundaries, capability negotiation and design principles. ### [MCP base protocol](https://modelcontextprotocol.io/specification/2026-07-28/basic) Authority: [Justin Spahr-Summers](https://mudpie.ai/authors/justin-spahr-summers/) **TL;DR:** MCP is a stateful JSON-RPC protocol with initialization, negotiated capabilities, notifications and explicit lifecycle behavior. **What you'll read:** Message shapes, request and notification rules, initialization, capability exchange and the primitives every SDK must preserve. ### [MCP transports](https://modelcontextprotocol.io/specification/2026-07-28/basic/transports) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** Local stdio and remote Streamable HTTP solve different connection problems while carrying the same protocol messages. **What you'll read:** Transport requirements, session handling, origin checks, protocol-version headers and why old HTTP+SSE assumptions no longer define remote MCP. ### [MCP server primitives](https://modelcontextprotocol.io/specification/2026-07-28/server) Authority: [Justin Spahr-Summers](https://mudpie.ai/authors/justin-spahr-summers/) **TL;DR:** Servers can expose prompts, resources and tools; treating all three as interchangeable functions throws away useful protocol semantics. **What you'll read:** The server capability model and the distinct jobs of prompts, resources, tools and utility features. ## Build reliable servers A production server is mostly boundaries: transport, authorization, validation, cancellation, discovery and recovery. ### [Build an MCP server](https://modelcontextprotocol.io/docs/2026-07-28/develop/build-server) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** Start with one useful capability and a supported SDK, then prove the server through an inspector or real host before adding breadth. **What you'll read:** A working server walkthrough and the minimum concepts needed to expose tools without hand-rolling the protocol. ### [MCP tools specification](https://modelcontextprotocol.io/specification/2026-07-28/server/tools) Authority: [Justin Spahr-Summers](https://mudpie.ai/authors/justin-spahr-summers/) **TL;DR:** Tool schemas describe inputs and outputs, but the client still owns consent, presentation and safety around invocation. **What you'll read:** Discovery and call methods, schema contracts, structured output, errors, annotations and the user-interaction principles around tools. ### [MCP authorization](https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** Remote MCP authorization builds on OAuth while keeping resource servers, authorization servers, clients and protected resources explicit. **What you'll read:** Discovery, metadata, audience binding, protected-resource behavior and the flow a conforming remote client should follow. ### [MCP authorization security considerations](https://modelcontextprotocol.io/specification/2026-07-28/basic/authorization/security-considerations) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** Token passthrough, confused deputies, open redirects and excessive scopes turn a convenient connector into an account-compromise path. **What you'll read:** Concrete attack classes and mitigations for audience validation, consent, redirect handling, least privilege and secure token use. ### [MCP client best practices](https://modelcontextprotocol.io/docs/2026-07-28/develop/clients/client-best-practices) Authority: [Justin Spahr-Summers](https://mudpie.ai/authors/justin-spahr-summers/) **TL;DR:** Clients must treat servers as untrusted capability providers, preserve user control and make tool state and failures understandable. **What you'll read:** Operational guidance for permissions, confirmations, capability changes, errors, timeouts, cancellation and safe user experience. ## Learn from the maintainers The spec tells you what MCP is. The roadmap, registry and governance posts explain why it changes and who is accountable for that change. ### [MCP's first anniversary](https://blog.modelcontextprotocol.io/posts/2025-11-25-first-mcp-anniversary/) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** MCP's first year moved the protocol from an Anthropic launch to broad infrastructure with a community process and a much larger compatibility burden. **What you'll read:** A maintainer retrospective on adoption, specification evolution, ecosystem growth and the problems still unsolved after year one. ### [Introducing the MCP Registry](https://blog.modelcontextprotocol.io/posts/2025-09-08-mcp-registry-preview/) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** A shared registry standardizes discovery metadata; it does not certify server quality, safety, uptime or real tool execution. **What you'll read:** Why the registry exists, how publishers and aggregators interact with it, and what canonical metadata can and cannot prove. ### [Governance and stewardship](https://modelcontextprotocol.io/community/governance) Authority: [Justin Spahr-Summers](https://mudpie.ai/authors/justin-spahr-summers/) **TL;DR:** MCP uses named maintainers, decision roles and public processes rather than letting protocol direction hide inside one vendor's product roadmap. **What you'll read:** Maintainer roles, decision authority, working groups, escalation and the public structure behind specification changes. ### [The MCP roadmap](https://blog.modelcontextprotocol.io/posts/2026-mcp-roadmap/) Authority: [David Soria Parra](https://mudpie.ai/authors/david-soria-parra/) **TL;DR:** The roadmap identifies the protocol's active pressure points so builders can avoid treating current gaps as permanent architecture. **What you'll read:** Maintainer priorities across transport, authorization, discovery, enterprise use, agents and extension mechanisms. ### [MCP joins the Agentic AI Foundation](https://blog.modelcontextprotocol.io/posts/2025-12-09-mcp-joins-agentic-ai-foundation/) Authority: [Anthropic](https://mudpie.ai/authors/anthropic/) **TL;DR:** Moving stewardship into a neutral foundation makes MCP less dependent on Anthropic while creating a formal home for ecosystem governance. **What you'll read:** The institutional transition, founding projects and what neutral stewardship changes—and does not change—for implementers.