modelcontextprotocol/modelcontextprotocol

The Model Context Protocol (MCP) is a category-defining initiative launched by Anthropic to solve the 'M x N' integration problem (M models needing to connect to N data sources). It is essentially the 'Language Server Protocol (LSP) for LLMs.' With 7.8k stars and a massive velocity of 1.04/hr, it has achieved instant escape velocity. Its defensibility is derived from network effects: as more 'servers' (connectors for Slack, Postgres, GitHub) and 'hosts' (Claude Desktop, IDEs, agent frameworks) adopt the protocol, the switching cost for the entire industry increases. While Anthropic (a frontier lab) created it, the risk to the project from other frontier labs is 'low' because it is an open standard designed to benefit the entire ecosystem; OpenAI or Google would more likely integrate with it than try to kill it, as it offloads the burden of building custom connectors. The primary risk is a competing standard from a Microsoft-OpenAI alliance, but MCP's head start and open-source nature give it a significant moat. The high platform domination risk reflects the fact that while the protocol is open, the most powerful 'hosts' will likely be the major LLM providers themselves.