Model Context Protocol (MCP)

Model Context Protocol (MCP) is an open standard that defines how AI models connect to external tools, data sources, and services — providing a common interface that lets agents and assistants access files, databases, APIs, and local software without requiring a custom integration for each capability.

Why a standard protocol matters

Before MCP, each application that connected a language model to external capabilities built its own integration layer. A coding assistant that needed to read files, query a database, and call an API required three separate, application-specific integrations. When the model changed or the external service changed, each integration needed to be updated independently. MCP addresses this fragmentation by defining a standard interface: tools and data sources implement the MCP server specification once, and any MCP-compatible client — a model host, an agent framework, or a developer tool — can use them without additional integration work.

How MCP is structured

MCP defines three core primitives that servers expose to clients. Tools are functions the model can invoke — calling an API, writing a file, querying a database — that have defined input schemas and return structured results. Resources are readable data sources — files, database records, configuration — that the model can access as context. Prompts are reusable prompt templates that surface server-specific instructions to the model. An MCP server implements some or all of these primitives for a specific capability domain; an MCP client connects to servers and makes their primitives available to the model at inference time.

Relationship to agentic AI

MCP is particularly significant for agentic AI systems, where a model needs to interact with many external capabilities over the course of completing a task. Rather than requiring agent developers to maintain a bespoke integration for every tool their agent uses, MCP provides a plugin-like model where capabilities are packaged as servers that any compatible agent can load. This changes the development economics: a database MCP server built once works with any agent framework that supports the protocol, and agent developers can compose capabilities from multiple servers rather than building each integration themselves.

Model Context Protocol (MCP) — FAQ

Is MCP specific to any particular AI model or provider?

MCP is designed to be model-agnostic. The protocol defines how a client and server communicate — what messages they exchange and what data structures they use — without specifying which model processes the results. Any model host that implements the MCP client specification can use MCP servers, and any developer can build an MCP server for any capability without targeting a specific model.

What is the difference between MCP and an AI agent's built-in tool calling?

Built-in tool calling is a capability of a specific model — the model is trained to recognize when to call a function and formats the call according to that model's schema. MCP is a transport and interface standard that sits above the model layer: it defines how tool definitions are communicated to clients, how tool calls are routed to servers, and how results are returned. MCP can be used with models that have native tool calling as the transport mechanism, standardizing what gets passed through the model's tool-calling interface.