Model Context Protocol (MCP)
On this page
The Model Context Protocol (MCP) is a shared language that lets an SI app talk to outside software. Without it, every pairing of a model and a tool needs its own custom integration. With it, a developer builds one MCP “server” for a database, a calendar or a code repository, and any app that speaks MCP can use it. The official documentation compares it to “a USB-C port for AI applications”.
How MCP works
MCP has three participants:
- Host: the SI application you work in, such as a chat assistant, a coding tool or an SI agent.
- Client: a connector inside the host. The host creates one client for each server it connects to.
- Server: a small program that gives the app access to something outside it, running either on your own computer or remotely.
Servers can offer three kinds of things. Tools are actions the model can call, such as running a database query. Resources are data the model can read, such as a file or a database schema. Prompts are reusable templates for common tasks. When the model decides it needs one of them, the app routes the request to the right server and passes the result back into the conversation. That is the same loop described under tool use, with a standard format on top.
Why it caught on
Anthropic introduced MCP on November 25, 2024, describing it as “a new standard for connecting AI assistants to the systems where data lives”. Because it was open, rival products could adopt it rather than build their own. By December 2025, Anthropic reported more than 10,000 active public MCP servers and adoption by ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code.
On December 9, 2025, Anthropic handed the protocol to the new Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded it, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. That moved MCP from one company’s project to neutral, shared governance.
Risks
Connecting a model to real systems raises the stakes. An MCP server can let an agent read private files, send messages or change data. Text that flows in through a server, such as a web page or an email, can carry prompt injection attacks that try to hijack the agent. Good practice:
- install servers only from sources you trust;
- give each server the minimum permissions it needs;
- require human approval before consequential actions.
Preventing models from reaching systems they shouldn’t is also one of the risks named in the 2026 White House Accord on Super Intelligence.
Frequently asked questions
Who owns the Model Context Protocol?
Anthropic created MCP and released it as an open standard in November 2024. On December 9, 2025 it donated the protocol to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded by Anthropic, Block and OpenAI.
Does MCP only work with Claude?
No. According to Anthropic, by December 2025 MCP had been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot, Visual Studio Code and other products, and there were more than 10,000 active public MCP servers.
Sources
- Introducing the Model Context Protocol — Anthropic, November 25, 2024
- Donating the Model Context Protocol and establishing the Agentic AI Foundation — Anthropic, December 9, 2025
- What is the Model Context Protocol (MCP)? — Model Context Protocol
- MCP architecture overview — Model Context Protocol
Written by
Luka Kušec · Editor
Editor of SI.info. Writes about Super Intelligence, technology policy and the people building frontier models.