modelcontextprotocol/servers

The Model Context Protocol (MCP) is rapidly becoming the 'Language Server Protocol' (LSP) for the LLM era. With over 83,000 stars and 10,000 forks in a relatively short timeframe, it represents a massive shift in how AI models access external data. The defensibility is a 10 not because the code is complex, but because of the powerful network effect: as more developers build MCP servers (Postgres, Slack, GitHub, etc.), more LLM clients (Claude, Cursor, Zed) adopt the protocol, creating a double-sided market lock-in. Anthropic's backing gives it institutional weight, but the open nature of the repo encourages community ownership. Frontier risk is low because this protocol solves a fragmentation problem that benefits all labs; while OpenAI or Google might introduce competing standards, the sheer momentum of MCP makes it the likely winner for developer-facing integrations. Platform domination risk is medium because while the protocol is open, the ecosystem is currently centered around Anthropic's Claude ecosystem, though this is diversifying rapidly into IDEs and other agentic frameworks. Displacement is unlikely because, like LSP, once the industry settles on a standard for tool-use, the cost of switching every integration to a new protocol is prohibitively high.