MCP (Model Context Protocol)
What is MCP (Model Context Protocol)?
MCP (Model Context Protocol) is an open standard that lets AI assistants connect to outside tools and data in the same way, like one plug socket for every AI assistant.
Anthropic released MCP as an open standard in November 2024. In December 2025 it became a founding project of the Agentic AI Foundation under the Linux Foundation, with OpenAI, Google and Microsoft taking part. A service that builds an MCP server once can be reached by every assistant that supports MCP, with no separate work for each one.
MCP (Model Context Protocol) in a shop
The Eurekanda workspace has an MCP server: your AI assistant prepares videos and posts as drafts, and you still press publish.
How does MCP (Model Context Protocol) work?
- 1A service (sales software, Google Sheets, a media library…) runs an MCP server describing the tools it offers: read orders, create drafts, search products.
- 2An AI assistant that supports MCP (such as Claude, ChatGPT or Cursor) connects to that server once you allow it.
- 3When you give it a job, the assistant picks the right tool and calls it over MCP; the service does the work and sends back the result.
- 4One MCP server works for every assistant that supports MCP, like a standard socket: no separate wiring for each.
What a shop or café can use it for
- Let an AI assistant read your stock sheet in Google Sheets to answer "is size M in stock?".
- Let the assistant prepare posts and videos as drafts in the workspace; you still press publish.
- A café connects its assistant to the booking calendar to check for a free table at 7 pm.
What to watch out for
- Each MCP server gives AI a real permission: grant read-only when reading is enough, and never delete or refund rights.
- Install MCP servers only from sources you trust; an unknown server can read data or plant instructions that trick the AI.
- MCP is a way to connect, not a way to make AI more accurate: wrong data still means wrong answers.
Quick questions
An API is each program's own doorway, and every assistant needs its own wiring. MCP is a shared standard on top: a program that runs an MCP server (usually calling its API underneath) works with every assistant that supports MCP.
No, if the service already offers an MCP server: you paste its address and access key into your AI assistant. Building an MCP server for your own software needs a technical person.
As safe as the permissions you grant. Use servers from trusted sources, grant as little as possible, and keep a person approving anything involving money.
Sources
The dated facts on this page come from the original sources below.
Keep learning beyond the glossary
Read next
- Agents and automationTool useTool use (function calling) is AI calling an outside function, such as checking stock, pricing delivery or creating an order, instead of only writing text.
- Agents and automationAPIAn API is a doorway that lets two pieces of software talk to each other by agreed rules, for example so AI can read orders or post to social media.
- Agents and automationAI agentAn AI agent is AI given a goal that takes several steps on its own to reach it: looking things up, using tools, checking the result and reporting back.