The Model Context Protocol Standard

About this paper

No specification in the history of APIs has been adopted this fast. Twenty months after its first dated version, MCP’s GitHub organization holds 42 repositories and 183,460 stars, ten official SDKs, and a servers repository with 89,090 stars — an order of magnitude more than the OpenAPI Specification after nine years.

API Evangelist has argued all year that MCP is last-mile plumbing: worth building when it sits on a real versioned API, and pure overhead when it is a costume over one. This is the first time numbers can be put against that. 3,060 providers carry an MCP artifact; 672 publish a server you can actually reach — and 1,879 are tool lists API Evangelist derived because no server exists. 64,537 tools are described, the median server exposes 11, and the largest exposes 588. A server handing an agent 588 tool definitions has not made an API usable; it has moved the discovery problem into the context window. The full report has the provenance separation, the tool-bloat distribution, the governance numbers (42.1% top-ten concentration — the least concentrated community in this series), the unfinished discovery story, and why MCP amplifies an authorization gap it did not cause.

What's inside

1. Executive summary — 3,060 artifacts, 672 reachable servers
2. What MCP is, and what it is not
3. The provenance separation — read before any number
4. The adoption ladder
5. The specification, and the shape of the organization
6. Tool bloat — the quality finding nobody measures
7. Discovery, the registry, and Agent Cards
8. Authorization — the gap MCP inherits and amplifies
9. The scoring caveat, and what the numbers do not say
10. What would have to be true
11. The investable thesis
12. Where this is going

What you get for $500.00