The Model Context Protocol, or MCP, is an open standard that defines how an AI application talks to an external system: how it discovers what is available, how it calls a tool, and how results come back. Anthropic published it and released it as an open specification with reference implementations.
MCP describes a client and a server exchanging JSON-RPC messages. The host is the application a person is actually using. Inside it, a client holds one connection to one server, and the server is the thing that knows how to talk to a particular system: a database, a file system, a CRM, a search index.
What the standard fixes is the shape of the conversation. How a server declares its tools, how arguments are described, how a call is made, how errors and results are returned, how a server can ask the client for something back. Because that shape is fixed, an agent does not need bespoke code per integration, and a vendor does not need to ship a different connector for every AI product on the market.
What the standard deliberately does not fix is what any of it means. MCP will not tell you
whether a tool called search_contacts returns verified emails or guesses, whether the
account behind it has quota left, or whether the vendor will still maintain the server next quarter.
Those are directory questions, not protocol questions, which is precisely the gap this site was
built to fill.
The specification is versioned by date and has been revised several times since first release, adding an authorization framework, a streamable HTTP transport alongside the original local transport, and richer tool output. Read the versioned spec rather than any blog post, including this one, for the current state of it.
Adoption is visible in this directory's own data rather than in anyone's press release. 144 of 293 go to market vendors ship a first party server today, and the auth notes on those entries name Claude, ChatGPT and Cursor as the clients they document sign in flows for.
GTM runs on a stack of systems that do not talk to each other: a CRM, an enrichment vendor, a sequencer, a call recorder, a warehouse. The reason a GTM engineer spends their week on plumbing is that every pair of those systems needs its own bridge. A protocol that makes each system speak once, to anything, is the first credible attack on that problem, and the reason it matters to measure which vendors have actually adopted it.
Every number on this page is generated from directory.json at build time and carries the date it was baked: 2026-08-25. Nothing is typed by hand, nothing is rounded, and nothing is estimated. The underlying data is published in full. Where the honest answer is a zero, the zero is printed.