What MCP is, in one sentence
The Model Context Protocol (MCP) is an open standard that defines how an AI application talks to external tools and data. Think of it as a universal adapter: instead of writing a custom connector for every combination of AI app and external system, you build one MCP server per system, and any MCP-compatible app can use it. It was introduced by Anthropic in November 2024 and has since gained SDKs and servers across the ecosystem.
The problem it solves
AI assistants are far more useful when they can read your files, query a database, hit an API, or run a command — but wiring those connections up has historically been painful. Each app reinvented its own plugin format, so a connector built for one assistant did not work in another. With M applications and N tools, the industry faced an M×N integration explosion. MCP collapses that into M+N: write one server for a tool, and every client speaks to it the same way.
The core architecture
MCP has three roles. The host is the user-facing AI application — a desktop chat client, an IDE extension, or an agent runtime. Inside the host, each connection is managed by an MCP client, a small component that maintains a one-to-one link with a server. The MCP server is a standalone program that exposes capabilities. Communication runs over a transport (commonly standard input/output for local servers, or HTTP with server-sent events for remote ones) using JSON-RPC 2.0 messages. The host can spin up many clients at once, each bound to a different server, giving the assistant access to several systems simultaneously.
The three things a server exposes
An MCP server can offer three kinds of capability. Tools are functions the model can choose to call — for example, “create a calendar event” or “run this SQL query”; these are model-controlled actions. Resources are read-only data the host can load into context, such as a file, a document, or a database row; these are application-controlled. Prompts are reusable, parameterised templates the user can invoke, like a “summarise this PR” command. Together these cover the three ways an AI needs to interact with the outside world: act on it, read from it, and reuse curated instructions.
How a typical exchange flows
When a host connects to a server, the two negotiate capabilities in an initialisation
handshake, agreeing on protocol version and which features each side supports. The host then
asks the server to list its tools, resources, and prompts. During a conversation, when the
model decides it needs a tool, the client sends a tools/call request; the server runs the
work and returns a structured result, which the host feeds back into the model. Because the
flow is standardised, the same model can use a GitHub server, a Postgres server, and a
filesystem server interchangeably.
Why MCP matters for agentic AI
MCP is foundational to the move from chatbots to agents. An agent that can plan and act needs a reliable, secure, well-described set of tools — and MCP gives it exactly that, with typed inputs, declared capabilities, and a clear permission boundary at the server. As the ecosystem of public and private servers grows, building a capable AI assistant increasingly means choosing which MCP servers to connect rather than writing integrations from scratch. That shift — from custom glue to a shared protocol — is why MCP is treated as the connective tissue of the agentic era.