MCP (Model Context Protocol)

Overview

MCP (Model Context Protocol) is an open-source standard that unifies how AI applications connect to external systems. The official documentation compares it to a USB-C port for AI applications. It lets AI applications such as Claude or ChatGPT access data sources like local files and databases, tools like search engines and calculators, and workflows like specialized prompts to carry out tasks.

It is split into MCP servers, which expose data and tools, and client applications, which connect to those servers. Servers are usually built with the official MCP SDKs for C#, Go, Java, Kotlin, PHP, Python, Ruby, Rust, Swift and TypeScript. Published servers can be found in the MCP Registry, and the GitHub repository has reference implementations for file operations, Git, web content fetching, knowledge graph-based memory and more.

It suits developers who need to attach multiple data sources and tools while building AI applications or agents. Once built, an integration can be shared across multiple clients that support MCP. The reference servers in the repository are educational examples, not meant for production, so developers must assess security requirements against their own threat model and put safeguards in place.

What it is used for

• Let an agent access Google Calendar or Notion so it works like a personalized assistant.
• Connect an enterprise chatbot to multiple databases across the organization so users can analyze data through chat.
• Have Claude Code build a web app from a Figma design.
• Have an AI model create a 3D design in Blender and print it on a 3D printer.
• Expose your own data and tools as MCP servers and reuse them across multiple AI clients.

Pricing and license

Open-source standard

Getting started

1. Read the documentation on MCP's core concepts and Architecture on the official site (modelcontextprotocol.io).
2. If you need an existing published server, look for it in the MCP Registry.
3. To build your own, choose the official MCP SDK for your language.
4. Implement a server or client using the reference servers on GitHub (Everything, Filesystem, Git and others) as examples.
5. Review the Security section of the official documentation and apply its security considerations and recommended practices.

Links

Official site ↗Repository ↗

Related terms and companies

MCP AI agent Tool use Function Calling LLM Agent memory

Recent mentions in briefings

• Uber shares design of 'MCP Gateway' that wraps existing APIs as MCP tools, running 800 MCP servers and 5,000 tools (community report) 2026-10-12
• Logpresso releases an integration app for analyzing ChatGPT Enterprise usage records in its SIEM (reported) 2026-10-12
• GitHub updates Copilot for JetBrains with default model and MCP server controls and improved chat navigation 2026-10-11

Sources

• What is the Model Context Protocol (MCP)? - Model Context Protocol
• modelcontextprotocol/servers README

Compiled from public sources (official pages, READMEs). Pricing and features change, so check the official site.