EVAL #009: MCP Hit 10,000 Servers. Is It Actually Ready for Production?

EVAL #009: MCP Hit 10,000 Servers. Is It Actually Ready for Production? By Ultra Dune | EVAL — The AI Tooling Intelligence Report | April 7, 2026 Every tool in your stack shipped an update this week. vLLM, PyTorch, TensorRT-LLM, Ollama, Transformers, LangChain, LlamaIndex, Qdrant — all of them. The common thread isn't a model. It's not a GPU. It's a protocol. The Model Context Protocol hit 10,000 registered servers this week. Anthropic's MCP v1.1 landed with OAuth 2.1, streamable HTTP transport, and tool annotations. LlamaIndex shipped MCP integration. Open WebUI 0.6 made it a first-class citizen. Pydantic AI 1.0 launched with native MCP client support. Cursor 1.0 leans on it. Even OpenAI — the company that spent months pretending MCP didn't exist — now supports it in the Agents SDK. MCP is no longer a bet. It's infrastructure. But infrastructure that grew this fast deserves scrutiny. Let's evaluate it properly. The Eval: MCP at Scale — Protocol, Promise, and the Production Gap What MC