modelcontextprotocol/modelcontextprotocol

Model Context Protocol is a de facto standard (7.7k stars, 1.4k forks, 560 days old, active governance) that solves the critical infrastructure problem of how AI models interact with external systems in a vendor-agnostic way. It's ecosystem-forming: Anthropic, OpenAI, and third-party developers are building servers and clients around this standard. The specification itself is the moat—not the implementation. Once a standard gains this level of adoption and institutional backing (explicit Anthropic sponsorship), it becomes strategically valuable infrastructure that competitors integrate with rather than replace. Frontier labs won't build competing protocols; they'll build MCP servers and clients. The low frontier risk reflects that this enables their products rather than competing with them. The novelty is novel_combination: JSON-RPC over stdio with a resource/tool/prompt abstraction is straightforward, but framing it as a unified protocol for AI grounding is the contribution. Defensibility is very high because (1) adoption network effects are already established, (2) ecosystem lock-in through server/client implementations, (3) specification standardization raises switching costs, (4) backed by major labs. Risk of displacement is near-zero; risk of expansion/evolution is high but extends rather than obsolesces the project.