Glossary
Model Context Protocol
An open protocol for connecting an AI application to external tools, data, and workflows. The specification’s current version, read November 6, 2026, is 2026-07-28.
The Model Context Protocol (MCP) is an open protocol for connecting an AI application to external systems. The introductory page, read November 6, 2026, describes it as a standard way for applications such as Claude or ChatGPT to reach data sources, tools, and workflows. The same page compares MCP to a USB-C port: one connection pattern, many devices. It names support in Claude, ChatGPT, Visual Studio Code, Cursor, and MCPJam.
The specification for revision 2026-07-28 says MCP uses JSON-RPC so hosts, clients, and servers can share context, expose tools, and build workflows. Servers can offer resources, prompts, and tools. The versioning page for that revision says the current protocol version is 2026-07-28, and that a version string marks the last date of a backwards-incompatible change.
MCP is a tool use path. The harness decides which servers are connected and what the agent may call. A skill served over MCP is a separate extension of the protocol. The file format those skills use is an agent skill. The evaluation guide, including NVIDIA’s November 4, 2026 example of an assistant connected through MCP, is How do you evaluate an AI agent harness?.