1. Home
  2. Glossary
  3. Model Context Protocol (MCP)
AI glossary · Agents & automation

Model Context Protocol (MCP)

Model Context Protocol (MCP): The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, that lets AI assistants connect to external tools, data sources, and services through a common interface instead of custom one-off integrations.

Before MCP, every AI product needed its own connector for every service: one integration for Google Drive, another for Slack, another for your database, rebuilt for each assistant. MCP defines a shared language. A developer builds one MCP server for a service, exposing its tools and data, and any MCP-compatible assistant can use it. Think of it as a universal plug for AI tools.

An MCP server can expose three kinds of things: tools the model can call (search tickets, create a record), resources it can read (files, database rows), and reusable prompts. The assistant, acting as the MCP client, discovers what is available and calls it when useful. Since its release, MCP has been adopted across the major assistants and developer tools, and thousands of community and vendor servers exist for common business software.

For a non-developer, MCP mostly shows up as a settings menu of connectors you can switch on: your calendar, your file storage, your CRM, your project tracker. The questions to ask before enabling one are the same as for any integration. What can it read? What can it change? Who published it? A connector from an unknown source can read whatever it is granted, so treat it like installing software.

Example at work

An account manager enables an MCP connector for her company's CRM inside her AI assistant. Now she can ask "summarize my open deals over $50k with no activity in two weeks" and get a real answer from live data. Her admin configured the connector read-only, so the assistant cannot edit a deal record.

Why it matters

MCP is why AI assistants are quickly becoming able to reach your actual work systems. Knowing what it is helps you understand what a connector can do, ask the right permission questions, and spot when a vendor's "integration" is a standard MCP server you could use elsewhere.

Related terms