This is a Hacker News thread asking "Who is using MCP in production?" The MODEL CONTEXT PROTOCOL (MCP) is Anthropic's open standard for connecting AI models to external tools and data sources. The thread contains ~50 substantive production use cases from practitioners. **Actual uses reported:** - **Non-technical user access**: SaaS companies (Tredict, adityapatadia's firm) deploy MCP so customers can connect their data to Claude/ChatGPT via OAuth flows without CLI knowledge—this appears to be the strongest legitimate use case. - **Enterprise tool integration**: Notion, Adidas, and others wrap internal APIs with MCP to give agents discoverable, sandboxed access to company systems rather than raw credentials. - **Domain-specific aggregation**: Log/metrics correlation across legacy systems (5ersi), voice agents for appointment scheduling (MitziMoto), Jira/Figma integrations for code work. - **Personal/hobby projects**: Language learning (navigate8310's Chinese textbook MCP), social media archive search, home automation, PDF markup workflows. **Recurring tension in thread**: Some argue CLIs are more token-efficient and already have training data; others counter that MCPs provide schema discovery, auth handling, and capability boundaries that CLIs don't. MCP appears strongest when the end user is non-technical or when you need to sandbox an agent's permissions strictly. The "is this just a wrapper?" criticism comes up repeatedly—fairly, since several respondents basically wrap REST APIs. The thread shows MCP is quietly used in production by mid-market SaaS and enterprises, though not at the scale early hype suggested.