Black October, explainedAppsGuidesAI DictionarySmall FixesOn RecordAboutFollow the lab

Model Context Protocol

Also called MCP, MCP server, connectors

MCP is a shared standard that lets any AI app plug into any tool or data source the same way, like one charging port for everything.

Using AITechnical

Products, laws and numbers here change fast. Checked October 2026.

In plain words

AI apps get far more useful when they can reach your files, calendar, email or company database. Before MCP, every pairing needed its own custom connection. Three AI apps and four services meant twelve separate integrations, each built and maintained by someone.

The Model Context Protocol, or MCP, is an agreed way for the two sides to talk. A service builds one MCP server that describes what it offers. Any AI app that speaks MCP can then connect to it. Twelve custom links become seven standard ones.

Anthropic released MCP as an open standard in November 2024. It was quickly adopted by other AI companies, and many apps now call these plug-ins "connectors". In settings, adding a connector is usually adding an MCP server.

Everyday example

One charger for every phone

Not long ago, every phone brand had its own charger, and every drawer had a tangle of cables. Then most devices moved to USB-C: one port, any charger. MCP does that for AI. A calendar service builds one MCP plug, and ChatGPT, Claude, a code editor or an agent can all use it.

Where the example breaks: A charger only carries power. An MCP connection carries data and actions, so connecting one is closer to giving an app a key. Only add servers you trust, and check what they can do.

Go deeper

How it actually works

MCP is a protocol: a fixed set of message formats both sides agree to use. It does not change the AI model, only how apps hand it tools and data.

  1. Hosts, clients and servers

    The AI app (the host) opens one MCP client connection per MCP server. A server is a program that offers context or actions. It can run on your own computer, or remotely over the internet.

  2. Three things a server can offer

    Tools: actions the model can call, like search files or create event (tool use). Resources: data to read, like a document or a database schema. Prompts: ready-made templates for common tasks.

  3. Discover, then call

    The client asks the server for its list of tools, each with a name, description and input format. The app shows these to the model. When the model requests one, the client sends a tools/call message and returns the result to the model.

  4. Under the hood

    Messages use JSON-RPC 2.0, a simple request-and-response format. Local servers talk over standard input and output; remote ones over HTTP, with sign-in usually handled by OAuth, a standard way to grant access without sharing your password.

Released
25 November 2024, by Anthropic, as an open-source standard
Adopted by
ChatGPT, Gemini, Microsoft Copilot, Visual Studio Code, Cursor and others, according to Anthropic in December 2025
Who runs it now
Since December 2025, the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block and OpenAI
Scale
More than 10,000 active public MCP servers as of December 2025

MCP's own site uses the same picture: "Think of MCP like a USB-C port for AI applications."

Questions people ask

What is an MCP server?

A small program that exposes a service, such as a calendar, a code repository or a database, in the MCP format. It lists the actions and data it offers, and carries out requests from AI apps that connect to it. It can run locally or online.

Is MCP only for Claude?

No. Anthropic created it, but it is open and vendor-neutral. As of December 2025 it was supported in ChatGPT, Gemini, Microsoft Copilot, VS Code and Cursor, and in December 2025 it moved to the Linux Foundation's Agentic AI Foundation.

What is the difference between MCP and an API?

An API is any service's own doorway, each with its own design. MCP is one standard wrapper that AI apps understand. Many MCP servers simply translate between MCP and a service's existing API, so the AI app doesn't need custom code for each.

Is MCP safe?

It is as safe as the servers you connect. A server can read data and take actions, and text it returns can carry hidden instructions (prompt injection). Use servers from sources you trust, grant the least access needed, and keep approvals on.

What are connectors in ChatGPT or Claude?

Connectors are the apps' friendly name for links to outside services like Google Drive, Gmail or GitHub. Many are built on MCP, and some apps let you add your own by pasting an MCP server's address.

Further reading