MCP (Model Context Protocol)
In shortMCP (Model Context Protocol) is an open standard for connecting AI applications to external data, tools and workflows through one shared interface.
In plain words
Before MCP, each AI app needed its own custom link to each tool: one for the calendar, one for the files, one for the database. MCP replaces this with a shared protocol. The project's own introduction uses an image: think of MCP like a USB-C port for AI applications. An AI application that speaks MCP can connect to any MCP server in the same way.
Two roles matter:
- Client: the AI application, such as a chat app, a code editor or an agent.
- Server: a small program that exposes something to the client, such as local files, a database, a search engine or a set of ready prompts.
The official introduction names three kinds of things a server can offer: data sources (for example local files and databases), tools (for example search engines and calculators) and workflows (specialised prompts). The rules are written in a public specification that carries a date as its version. The introduction lists support in assistants such as Claude and ChatGPT and in developer tools such as Visual Studio Code and Cursor.
Why it matters
- Less glue code: Build a server once, and any MCP-capable app can use it.
- Choice: You can switch the AI app and keep your tools.
- Tool use at scale: MCP does not replace tool calling. It standardises how tools are described and found. An AI agent can then use many tools without custom code for each.
- Cost: Each connected server adds tool descriptions to the prompt. They use tokens and space in the context window. Connect only what you need.
- Security: A server can read data and take actions in your name. Install servers only from sources you trust, grant the narrowest access, and require approval for actions with side effects. Text that tools return can carry hidden instructions (prompt injection). The MCP documentation has a page on security best practices.
Example
You add a calendar server to your AI app. You ask: "When am I free on Thursday afternoon?" The app sees the calendar tools the server offers, asks the server for Thursday's events, and the model reads the list and answers. Later you swap to another AI app that also supports MCP, and the same calendar server works there too.
Common misconceptions
- "MCP is an AI model." It is a protocol, a set of rules for connecting apps and tools.
- "MCP servers are sandboxed and safe." Many run with your permissions. Treat them like software you install.
- "MCP belongs to one vendor's app." It is an open standard that several vendors and tools support.
Questions and answers
What does MCP stand for?
Model Context Protocol. It is an open standard for connecting AI applications to data sources, tools and workflows.
What is the difference between MCP and function calling?
Function calling is how a model asks for a tool. MCP is a shared protocol for how tools are offered to AI apps, so one server works with many clients.
Is it safe to install MCP servers?
Only from sources you trust. A server can access data and act in your name. Give it the least access it needs and review risky actions.
Read next
- Tool use (function calling)
Tool use, also called function calling, lets a model request an outside action, such as a search or a calculation, which an app runs and feeds back.
- AI agent
An AI agent is a system where a model plans, uses tools and checks results in a loop to reach a goal with little step-by-step guidance.
- Coding agent
A coding agent is an AI tool that reads a codebase, edits files and runs commands and tests in a loop to finish a programming task.
- Prompt injection
Prompt injection is an attack where text a model reads, such as a web page or file, contains instructions that make it act against its user.
- Context window
The context window is the maximum amount of text, counted in tokens, that a model can read and write in one request.