Model Context Protocol (MCP)
Also written MCP
The Model Context Protocol is an open standard that lets an AI application discover and call tools on an external server over a shared JSON-RPC interface. The client opens a connection, agrees a protocol revision with the server, asks what the server can do, and calls what comes back. Anthropic published it in November 2024.
What is the Model Context Protocol?
MCP is a wire protocol for connecting AI applications to external capabilities. It standardises the conversation, not the capability: any server that speaks it can be plugged into any client that speaks it.
Before MCP, every integration between an assistant and an external system was bespoke. Each vendor shipped its own plugin format, its own authentication story, and its own way of describing what a function did. MCP replaces that with one exchange that every participant implements once.
What does MCP actually standardise?
Three things: the transport a client connects over, the handshake that agrees a revision, and the method names for discovering and invoking capabilities. Everything above that is the server author's business.
The pieces a working connection touches are all defined by the specification: JSON-RPC 2.0 as the message format, initialize as the opening exchange, capabilities as the negotiated feature set, and tools/list as the discovery call. Tools are the most used capability, but the protocol also covers resources and prompts.
How do you know a server speaks it?
You connect and ask. A server that speaks MCP answers an initialize request with a protocol revision and a server name, then answers tools/list with an array of tool objects.
That is a real answer rather than a claim on a README, which is the distinction that matters when a directory tells you something is an MCP server. We attempted 53 endpoints between 20 July and 5 August 2026 and reached a usable tools/list on 29 of those attempts.
Example
On 5 August 2026 we connected to 14 public MCP servers in one run. Each one negotiated a revision, named itself, and returned its tool list: Context7 answered with 2 tools, Microsoft Learn with 3, and a game server called spacemolt with 213. The exchange was identical in all 14 cases, which is the entire point of a protocol.