MCP has rapidly become the de facto standard for tool calling in AI systems - not because it was perfectly designed, but because it was the first open protocol to gain real traction. In this talk, I detail how MCP evolved from an early standard to a widely adopted foundation across major LLM providers and thousands of public servers.

With explosive growth comes serious challenges: fragmented implementations, unresolved security risks such as tool poisoning and prompt injection, and context bloat as servers accumulate large tool definitions. The ecosystem continues to shift quickly - new features like asynchronous operations and programmatic tool calling have landed, and as of December 2025, MCP has moved under the Linux Foundation’s Agentic AI Foundation, bringing neutral governance from major industry contributors.

We will cut through the complexity to show what the protocol truly specifies, how it’s used in practice, and which features matter to build robust clients and servers as MCP enters its next phase.