An open-source standard created by Anthropic that gives AI models a universal way to connect to outside data sources and tools. Since its 2024 launch, it has been adopted by rivals including OpenAI and Google, and in 2025 was donated to the Linux Foundation.

Developers build MCP "servers" — plug-ins that let an AI assistant tap into a specific capability. For example, the Context7 MCP server feeds an AI agent up-to-date documentation for software libraries, so it writes code using current syntax instead of outdated patterns.

The Playwright MCP server lets an AI agent open a web browser, navigate pages, and read their content. A data reporter could use it to have an AI assistant inspect a government website and then write a custom scraper to pull public records — turning what used to be hours of coding into a conversation.

Where MCP connects an agent to outside data and tools (the what), agent skills teach it workflows and best practices (the how). The two often work together: an MCP server might connect an agent to live market data, while a skill instructs it on how to analyze that data and format a report.