Anthropic introduced it in November 2024, and on 9 December 2025 donated it to the Agentic AI Foundation, a fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI. The specification is published at modelcontextprotocol.io.
USB-C is a fair comparison. Before it, every device needed its own cable; after it, one connector worked across many devices. MCP does the same for AI tools, so that instead of every agent needing a custom connection to every system, both sides speak one protocol.
How does MCP work?
A tool or service that wants to be reachable runs an MCP server. The server describes the actions it offers, such as looking up a record, creating a draft or searching a folder, in a standard format. An agent that supports MCP acts as the client: it reads that list of actions and calls them when a task needs one.
MCP is a standard, so there's nothing to buy. What you choose is software that speaks it, on the agent side and on the tool side.
Why does MCP matter to you?
It widens what an agent can reach without custom development. If a tool you rely on offers an MCP server, any agent that supports MCP can connect to it, and connecting becomes a matter of configuration rather than a building project.
It also reduces lock-in. A connection built to a shared standard can be reused by another vendor's agent if you ever switch, which makes MCP support a reasonable thing to ask about when comparing agents.
An example
A potter who sells online keeps her stock in a shop system that offers an MCP server. Her agent supports MCP, so when she asks "how many of the large blue jugs are left, and which glazes are running low?", it calls the stock lookup directly and answers from live figures. Nobody had to write an integration between the two products.
How is MCP different from an API?
An API is a specific interface to a specific service. MCP is a common way of describing tools so that many different agents can use them. They usually work together: an MCP server sits in front of a service's API and presents it in the standard shape that agents understand.
Common questions
Is MCP only for Claude?
No. Although Anthropic created it, MCP is an open standard, and by the time of the December 2025 donation it had been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code, among others. Governance now sits with the Agentic AI Foundation rather than any single company.
Do I need to understand MCP to use an AI agent?
No. Many agents use it behind the scenes without you seeing it. It becomes relevant when you want your agent to reach a particular tool and need to check whether that tool offers an MCP server.
Is MCP a security risk?
Every connection is a permission, and because MCP makes connections easy to add, it also makes it easy to grant access carelessly. Connect only servers from sources you trust, and give each one the narrowest access the task needs.