An open standard for how AI agents discover and invoke external tools and data sources.
The Model Context Protocol (MCP) is an open, vendor-neutral standard that defines a universal interface between AI agents (clients) and external capabilities (servers). Developed to solve the "N×M integration problem" in agentic development, MCP allows any compliant agent to connect to any compliant tool without custom integration code.
MCP servers expose a machine-readable schema describing available tools, their parameters, and authentication requirements. MCP clients (embedded in AI hosts like Claude or GPT-5) query these servers for available capabilities and invoke them using a standardized request format.
The protocol is often called "the HTTP of the agentic web" because—like HTTP—it provides a universal, layered abstraction that enables an ecosystem of interoperable components. Just as HTTP allowed any browser to load any website, MCP allows any agent to use any MCP-compliant tool.
Key features of MCP include OAuth 2.0 authentication for tool invocations, capability declaration schemas, sandboxed execution environments, and comprehensive audit logging.