Tools & agents
Model Context ProtocolMCP
Also known as: MCP
An open standard that lets AI applications plug into external tools and data through one common interface, instead of a custom integration for each.
The Model Context Protocol, or MCP, is a shared "language" that lets an AI assistant connect to other programs and data, your calendar, your files, a company database, without someone having to build a special bridge every single time. The easiest way to picture it: MCP is like a USB-C port for AI. One standard plug, and suddenly the assistant can connect to many different tools using the same connection instead of a unique cable for each one.
MCP is an open standard, introduced by Anthropic in November 2024 and now used across the industry, that defines how an AI application talks to external systems. Before a standard like this, every combination of "AI app" and "tool" needed its own custom integration. With M apps and N tools that is M × N pieces of glue code. A shared protocol turns that into M + N: each tool exposes an MCP interface once, and any MCP-capable app can use it.
What a tool exposes
An MCP connection can offer three kinds of things to the model: tools (actions the AI can take, such as sending an email or querying a database), resources (data the AI can read, such as documents), and prompts (reusable templates). The AI application is the client; the program exposing the capabilities is the server.
Why it matters
MCP is one of the building blocks that turns a plain language model into an AI agent that can actually do things in the real world, rather than only describe them.
MCP is a client-server protocol built on JSON-RPC 2.0 that standardises how a host application supplies context and capabilities to a language model. The architecture has three roles: a host (the AI application), one or more clients it spins up, and servers that expose capabilities. Servers advertise three primitives, tools (model-invocable functions, akin to function calling), resources (addressable read-only context), and prompts (parameterised templates), and communicate over pluggable transports, principally stdio for local processes and Streamable HTTP for remote ones.
The design goal is to collapse the M×N integration problem into M+N and to decouple capability providers from model vendors, so a server written once works with any compliant client. In practice MCP is a key layer in agentic systems: it gives the orchestration loop a uniform, discoverable way to call tools and pull in grounded data, complementing retrieval approaches such as RAG.