What Is an MCP Server? A Beginner’s Guide to the Model Context Protocol
Learn what an MCP server is, how hosts, clients, tools, resources and transports work, and how local and remote MCP differ in the current protocol.
Independent Model Context Protocol research
The protocol moved fast—and a lot of explanations did not. MCP Blog traces every important claim to a source and separates today’s stateless core from yesterday’s session-based model.
Stateless core. Self-contained requests. Discovery before invocation. Tasks as an extension.
Read the annotated change guideWhy this site exists
“MCP” no longer describes one fixed architecture. It describes a protocol that crossed a major compatibility boundary.
Before July 2026, an MCP connection began with initialization and capability negotiation. The current core is stateless: requests carry their own context, servers advertise operations through discovery, and formerly core features can live as extensions.
That distinction changes how you explain MCP, design a server, debug a client, and assess security. Every technical page here names the protocol version it covers.
Evidence before opinion
Foundations, architecture, migration, and security—each tied to a protocol revision and a public source trail.
Learn what an MCP server is, how hosts, clients, tools, resources and transports work, and how local and remote MCP differ in the current protocol.
Compare MCP with native function calling and tool use. Learn how they work together, where APIs fit, their security and latency trade-offs, and when to use each.
A practical guide to MCP's stateless 2026 specification, including per-request metadata, version discovery, multi-round trips, subscriptions, extensions, and legacy compatibility.
A version-aware explanation of MCP: the problem it solves, its architecture and capabilities, and how the stateless 2026 core differs from legacy MCP.
A version-aware explanation of MCP architecture, primitives, message patterns, transports, and trust boundaries under the 2026-07-28 specification.
A practical migration guide from MCP's initialize-and-session era to the stateless 2026-07-28 protocol, with compatibility and state-design patterns.
A technical guide to MCP security covering OAuth, prompt injection, tool poisoning, malicious servers, CVEs, supply-chain risks, and secure local and remote deployment.
Compare MCP stdio and current Streamable HTTP, and distinguish the 2024 HTTP+SSE and 2025 session-based wire models.
A primary-source timeline of MCP, from its pre-launch repository and November 2024 announcement through adoption, open governance, and the stateless 2026 core.
MCP does not replace REST, GraphQL, RPC, or vendor SDKs. Learn what it standardizes at the AI boundary and where direct APIs still win.
MCP connects AI applications to tools and context; A2A connects independent agents. Compare their roles, objects, and how they work together.
Our standard
Every technical guide states the specification revision it describes and labels compatibility-era details.
Our public source library includes specifications, commits, standards, advisories, research, and first-party announcements.
MCP Blog is not an official MCP property. Vendor claims are treated as claims, not proof of adoption or quality.
The protocol in motion
Specification and SDK history begin before the public launch.
A stateful, JSON-RPC-based protocol with tools, resources, prompts, and more.
Streamable HTTP and a protocol authorization framework arrive.
MCP joins the Agentic AI Foundation under the Linux Foundation.
Stateless core, discovery, MRTR, extensions, and explicit deprecations.
Check our work
The research corpus spans the protocol specification, repository history, RFCs, governance, SDKs, vendor adoption, security advisories, and academic work.
Open the source library