Model Context Protocol (MCP)

10 tools, 10 integrations. Every connection between an LLM and an external system needed its own adapter. Proprietary, fragile, not reusable. The Model Context Protocol (MCP) solves this: an open standard connecting AI models with any system.

USB for AI

Developed by Anthropic, released in November 2024. The architecture: client-server model with JSON-RPC 2.0 and three primitives. Prompts are pre-built instruction templates. Resources deliver structured data. Tools are executable functions the model can call directly.

One standard instead of a hundred proprietary connectors. Exactly like USB unified the peripheral landscape.

From 100 to 5,800 servers in eight months

November 2024: 100 MCP servers. Mid-2025: over 5,800. 97 million SDK downloads per month. OpenAI adopted MCP in March 2025, Google integrated it for Gemini, Microsoft for Copilot Studio. December 2025: transfer to the Linux Foundation.

What this means for businesses: AI agents that access SAP, Salesforce, or internal databases directly, through a standardized, auditable protocol. No more custom integrations.