MCP: a standard plug between AI and your tools
The Model Context Protocol (MCP) is an open protocol that defines a common way to connect an AI application to tools, files, databases or services. A tool exposed once through MCP can then be used by several compatible assistants.
Data as of October 22, 2026
Key figures
| Level | Intermediate |
|---|---|
| Origin | 2024 |
| Introduced by | Anthropic |
How it works in practice
Before MCP, every AI assistant had its own way of plugging into a tool: connecting one service to three assistants meant three integrations. MCP offers a shared interface; the official documentation likens it to a USB-C port for AI applications.
The specification is built around three roles:
- the host: the AI application in use, such as a desktop assistant or a code editor;
- the client: the connector inside the host that talks to a server;
- the server: the program that exposes capabilities.
A server can offer tools (actions the model can trigger through tool calling), resources (data to read) and prompts (instruction templates). Messages use the JSON-RPC 2.0 format, and the design takes inspiration from the Language Server Protocol used by code editors.
A concrete example
A support team installs an MCP server for its ticketing tool. From their AI assistant, a support agent asks: “Summarise this week’s open billing tickets and suggest a template reply.” The assistant calls the server’s search tool, reads the tickets, then drafts the reply. If the team moves to another MCP-compatible assistant, the same server still works.
Why it matters
For AI agents and agentic workflows, MCP removes the need to rebuild the same integrations for every provider: a well-built connector works everywhere. If you are picking a framework to build an agent, check whether it supports MCP; our frameworks page helps you compare them.
The protocol was launched by Anthropic and then handed to a neutral foundation, as explained in the announcement of the donation to the Agentic AI Foundation.
Common mistakes and myths
- “MCP replaces APIs.” An MCP server usually sits on top of an existing API: it gives assistants a standard way in, it does not replace it.
- “MCP belongs to Anthropic.” Anthropic created it, but since December 2025 the protocol has been hosted by the Agentic AI Foundation, under the Linux Foundation.
- “An MCP server is safe by default.” The specification points out that tools can run arbitrary code and requires user consent. An untrustworthy server or booby-trapped content can also be used for prompt injection, so install only servers you trust.
- “MCP is for agents to talk to each other.” That role belongs to the A2A protocol.
Related terms
- Tool calling (function calling) · The ability of an AI model to request that an external function be run with specific arguments, then use the result in its answer.
- AI agent · An AI that does more than answer: it plans and carries out a series of actions with tools to reach a goal you set.
- API · An interface that lets one program send requests to a service, such as an AI model, and get its responses back automatically.
- Agent2Agent protocol (A2A) · An open protocol that lets AI agents from different vendors discover each other, communicate and hand tasks to one another.
- Agentic workflow · A sequence of steps in which an AI model plans, uses tools, checks and revises its work instead of answering in a single pass.
- Prompt injection · An attack that slips hidden instructions into text an AI reads, to make it ignore its original instructions or act against its user.
Try it
Full glossary
Frequently asked questions
Who created MCP, and when?
Anthropic announced it on 25 November 2024 as an open, open-source standard. The protocol was designed by David Soria Parra and Justin Spahr-Summers.
Who governs MCP today?
On 9 December 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF), a directed fund hosted by the Linux Foundation and co-founded by Anthropic, Block and OpenAI. The foundation now hosts the protocol, alongside other projects such as A2A and AGENTS.md.
Does MCP only work with Claude?
No. The official documentation lists Claude, ChatGPT, Visual Studio Code and Cursor among compatible clients. Avoiding dependence on a single provider is the whole point of a standard.
How is it different from the A2A protocol?
MCP connects an AI application to tools and data. The A2A protocol is about communication between agents. The two are complementary and are now hosted by the same foundation.
Data sources
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.