Glossary
MCP (Model Context Protocol): a plain-English definition
The standard that lets AI tools reach your apps and files. The full explainer covers setup and safety in more depth.
Definition
MCP, the Model Context Protocol, is an open standard that lets AI applications connect to external data sources, tools and workflows in a common way.
Full explainer: What is MCP? covers this in more depth, with examples.
What it means at work
In plain terms, the Model Context Protocol docs call MCP an open-source standard for connecting AI applications to external systems. The Model Context Protocol docs compare MCP to a port that connects AI applications to external systems. In practice, it is how an assistant such as Claude or ChatGPT can be given access to things like your files, calendar or other business tools. The MCP documentation says it is an open protocol supported across a wide range of clients and servers.
What the sources say
- On connections, the MCP introduction says AI applications like Claude or ChatGPT can connect to data sources, such as local files and databases.
- On structure, the Model Context Protocol architecture page says MCP consists of two layers: a data layer and a transport layer.
- On oversight, the MCP tools specification says there SHOULD always be a human in the loop with the ability to deny tool invocations.
- An arXiv study of MCP builds a threat taxonomy encompassing 16 distinct threat scenarios across four major attacker types.
- That study describes MCP as an emerging open standard that defines a unified communication protocol between AI models and external tools or resources.
- Anthropic’s own definition matches. Its glossary calls MCP an open protocol that standardizes how applications provide context to LLMs.
An example
You connect an assistant to a shared document folder through an MCP-based connector. Now you can ask, ‘Find the latest signed vendor agreement and list the renewal date,’ and the assistant reads the folder instead of you uploading files by hand.
A common misconception
MCP is not itself an AI or a product you buy. It is a standard, and every connection is only as safe as the permissions you grant. Connect only what a task needs, prefer read-only access first, and check with IT before linking systems that hold client or employee data.
Sources
- Model Context Protocol, What is MCP?: “open-source standard for connecting AI applications to external systems”
- Model Context Protocol, What is MCP?: “Think of MCP like a USB-C port for AI applications”
- Model Context Protocol, What is MCP?: “MCP is an open protocol supported across a wide range of clients and servers”
- Model Context Protocol, What is MCP?: “AI applications like Claude or ChatGPT can connect to data sources”
- Model Context Protocol, Architecture overview: “MCP consists of two layers … Transport layer”
- Model Context Protocol, Specification: Tools: “there SHOULD always be a human in the loop with the ability to deny tool invocations”
- “Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions” (arXiv): “encompassing 16 distinct threat scenarios”
- “Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions” (arXiv): “an emerging open standard that defines a unified”
- Anthropic, Claude Platform Docs glossary: “an open protocol that standardizes how applications provide context to LLMs”
Frequently asked
Do I need to code to use MCP?
Usually not. Many assistants offer connectors you switch on in settings. Building a new connector is a developer task.