GlossaryAI & AutomationAdvanced

MCP (Model Context Protocol)

MCP (Model Context Protocol) is an open standard that lets AI applications connect to external tools, data sources and services through one consistent client–server interface.

Last updated May 28, 2026

Definition

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic, that defines how AI assistants and agents communicate with external systems — tools, databases, APIs, files and more — through a single uniform interface. It is often described as a 'USB-C port for AI': one standard connector instead of a custom integration for every data source.

How MCP works

An MCP server exposes capabilities (tools to call, resources to read, prompts to reuse). An MCP client — embedded in an AI app or agent — discovers those capabilities at runtime and invokes them on the model's behalf. Because the contract is standardized, the same server works across any compliant client, and the same client can talk to any compliant server.

Why it matters

Before MCP, every AI-to-tool connection was a bespoke integration. MCP turns those into reusable, composable building blocks, which makes AI systems far more extensible and portable across vendors.

  • Tools — functions the model can call (e.g. run a query, fetch a URL).
  • Resources — data the model can read (files, records, documents).
  • Prompts — reusable templates a server can offer to clients.

Examples

1

An MCP server that exposes a company knowledge base to an AI assistant

2

Connecting a coding agent to a Git repository through an MCP server

3

A browser-automation MCP server that lets an agent scrape pages on demand

Common Use Cases

Giving AI agents secure, scoped access to internal tools
Standardizing integrations across multiple AI clients
Building reusable connectors for data sources
Letting agents trigger real-world actions safely

Frequently Asked Questions

MCP was introduced by Anthropic as an open standard and has since been adopted across a wide range of AI tools and clients.
No. MCP is model-agnostic — any client that implements the protocol can talk to any compliant MCP server, regardless of the underlying model.
MCP servers control exactly which tools and resources they expose, and clients typically require explicit user approval before an agent runs sensitive actions.