When Anthropic released the MCP protocol in November 2024, most people reacted with "yet another company pushing its own standard." Less than a year later, OpenAI, Google, and Microsoft have all adopted it, making MCP the de facto standard in the AI field.
What problem does MCP solve?
Previously, every AI model needed a separate adapter to interface with each external tool. M models × N tools = M×N connectors. MCP acts like USB-C — a unified interface that handles all connections.
Technically, it reuses the message flow design of the Language Server Protocol (LSP), with JSON-RPC 2.0 on the transport layer. Three core primitives:
- Tools: Execute code
- Resources: Data access
- Prompts: Templated interactions
Why did it win?
The key turning point was the collective adoption by industry giants:
| Date | Event |
|---|---|
| March 2025 | OpenAI fully adopts MCP |
| May 2025 | Microsoft and GitHub join the steering committee |
| 2025 | Google DeepMind follows suit |
| November 2025 | Protocol update: async operations, statelessness, server identity |
| December 2025 | Donated to AAIF under the Linux Foundation |
Once the three giants were on board, the rest was natural diffusion. The community has built thousands of MCP Servers, and SDKs are available for all major programming languages.
Final destination
In December 2025, Anthropic donated MCP to the Agentic AI Foundation (AAIF) — a directed fund under the Linux Foundation, co-founded by Anthropic, Block, and OpenAI. From then on, MCP is no longer a company's standard but neutral infrastructure.
This path is somewhat reminiscent of Kubernetes' development — Google built it, donated it to CNCF, and it became an industry standard. The AI field has finally had its own "K8s moment."
Source: CocoLoop, The New Stack