Model Context Protocol (MCP)
MCP (Model Context Protocol) is an open standard for connecting AI applications to external systems such as databases, business tools and content repositories. Introduced by Anthropic in November 2024, MCP gives AI assistants and agents one common way to read data and trigger actions in other software, instead of a custom integration for every tool.
Model Context Protocol: definition and how it works
MCP follows a client-server architecture with three participants. The host is the AI application, such as an AI assistant or a code editor. For each system it connects to, the host creates an MCP client, which keeps a dedicated connection to an MCP server: the program that exposes that system’s data and actions. Messages are exchanged in JSON-RPC 2.0 (MCP architecture overview).
An MCP server offers three main building blocks: tools, the actions the AI can call; resources, the data it can read; and prompts, reusable interaction templates. The official documentation compares MCP to a USB-C port for AI applications: one standard connector instead of a different cable for every device. That is what makes MCP a building block of agentic AI, where an agent has to act inside several applications.
Since December 9, 2025, MCP has been governed by the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI, to which Anthropic donated the protocol. In the same announcement, Anthropic listed ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code among the products that had adopted MCP (Anthropic, December 2025). The specification is versioned on modelcontextprotocol.io; the current version is dated July 28, 2026.
MCP, API and MCP server: the differences
An API is the interface that one piece of software exposes, such as a REST or GraphQL API, with its own endpoints and authentication, and each AI tool needs custom code to use it. MCP does not replace APIs. An MCP server usually sits on top of an existing API and describes its actions in a standard format that any MCP-compatible AI application can discover and call.
| API | MCP | MCP server | |
|---|---|---|---|
| What it is | The interface of one system | A protocol shared by AI applications and systems | A program that exposes one system through MCP |
| Who calls it | Developers, through code written for that API | AI applications, through their MCP clients | MCP clients |
| Integration effort | One integration per tool and per system | Build once, usable by any compatible application | One server per system |
Model Context Protocol in practice
When a content management system exposes an MCP server, an AI assistant can search content, draft a page, update a field or prepare a translation from a plain-language request. What the assistant is allowed to do is decided outside the protocol.
The specification states that MCP cannot enforce its security principles at the protocol level, and requires hosts to obtain explicit user consent before invoking any tool (MCP specification, Security and Trust & Safety). The guardrails therefore stay in the system behind the server: the permissions of the account the AI uses, the content workflow that routes a draft to review, and the human validation before publication. Before connecting an assistant to a CMS, check which account the server uses, which actions it exposes (read only, or write and publish), and whether each action is logged.
Frequently asked questions
What does MCP stand for in AI?
In AI, MCP stands for Model Context Protocol, the open standard that connects AI applications to external data and tools. The “context” is the information a model needs to answer or act, such as a document, a database record or the result of an action. The acronym has other meanings in medicine and electronics, which have nothing to do with this protocol.
What is an MCP server?
An MCP server is a program that exposes the data and actions of one system, such as a CMS, a database or a ticketing tool, through the Model Context Protocol. It can run locally on the user’s machine or remotely as a service. Any MCP-compatible AI application can connect to it, discover the tools it offers and call them, within the permissions the server grants.
Does ChatGPT support MCP?
Yes. Anthropic listed ChatGPT among the products that had adopted MCP in its December 2025 announcement, alongside Cursor, Gemini, Microsoft Copilot and Visual Studio Code. The official MCP documentation also names Claude and ChatGPT as supported AI assistants.
Is MCP secure?
MCP defines security principles but cannot enforce them on its own. The specification requires explicit user consent before any tool runs and treats tool descriptions as untrusted unless they come from a trusted server. In practice, security depends on the implementation: authentication of the server, the permissions of the account it uses, human validation of sensitive actions and a log of what the AI did.
Resources that cover this term
Loading resources and related terms…