Agents
Model Context Protocol (MCP)
An open protocol introduced by Anthropic in 2024 that standardizes how AI assistants connect to external tools and data sources, similar to how USB-C standardizes physical connections.
The Model Context Protocol (MCP) is an open specification by Anthropic (released November 2024) that standardizes how AI applications expose tools, resources, and prompts to language models. Before MCP, every assistant had its own custom integration for each data source — gmail, GitHub, Slack, internal APIs. MCP defines a single client–server protocol so any compliant AI app can connect to any compliant tool server.
The protocol uses JSON-RPC over stdio, HTTP, or SSE. An MCP server exposes capabilities (e.g., “search Gmail”, “execute SQL”, “read local files”); an MCP client is the AI assistant (Claude Desktop, Cursor, Claude Code, ChatGPT) that consumes these capabilities. Tool authors write the server once; users wire it up in the assistant of their choice.
By 2026, MCP has been adopted by OpenAI, Google, and most agent frameworks. Hundreds of community servers exist for popular SaaS, databases, and developer tools. The specification continues to evolve: bidirectional sampling (servers can ask the client to run completions), structured roots (filesystem and project context), and elicitation (servers can ask the user follow-up questions through the client).
MCP is the dominant solution to the “AI integration sprawl” problem of 2023–2024 and underpins much of the agentic ecosystem.