Nick Cooper

Nick Cooper helped deliver OpenAI’s Apps SDK, which lets developers build interactive applications inside ChatGPT using the Model Context Protocol (MCP). He also contributed direction to MCP Apps, the collaborative effort to standardize interactive interfaces across MCP hosts, bringing experience from the Apps SDK into the shared specification.[2][3]

The November 21, 2025 MCP Apps proposal listed Cooper as a core maintainer among its authors and credited him, Alexei Christakis and Bryan Ashley with direction informed by their Apps SDK work. The effort also built on patterns pioneered by MCP-UI creators Ido Salomon and Liad Yosef, with Anthropic contributors helping steer the initiative and the UI Community Working Group shaping it through feedback and testing.[2]

The problem was practical: a tool could return chart data as JSON, but each client then needed its own rendering logic. More complex interactions, such as collecting related settings, compounded that burden. MCP Apps established a common way for tools to reference interfaces and for those interfaces to communicate with the host, reducing the fragmentation created by incompatible UI conventions.[2]

The collaborative design separates presentation from tool results. Tools reference predeclared ui:// resources, allowing hosts to fetch and review templates before execution and cache them separately from dynamic data. The host renders the interface in a sandboxed iframe, with bidirectional communication using MCP JSON-RPC over postMessage. The November proposal also called for meaningful text-only fallbacks so tools could serve hosts without UI support.[2][3]

MCP Apps became an official extension on January 26, 2026, enabling tools to return dashboards, forms and other interfaces directly in conversations. Cooper, identified in the announcement as an OpenAI Member of Technical Staff, described its foundations in MCP-UI and the ChatGPT Apps SDK and said: “We’re proud to support this new open standard and look forward to seeing what developers build with it as we grow the selection of apps available in ChatGPT.”[3]

At MCP Dev Summit North America 2026, Cooper argued that MCP and command-line interfaces were complementary. In the Agentic AI Foundation’s May 15 account of his talk, he put it plainly: “You clearly want both.” His formulation, “MCP is an API for AI,” emphasized designing interfaces around what a user wants the model to accomplish.[4]

Cooper illustrated that view with email. Send, receive and list seem like a simple API, but drafts, labels, folders, categorization and contact management can expand it into hundreds of technically valid operations that confuse a model. He argued for a CLI or MCP layer that exposes useful abstractions rather than simply handing the model the full underlying API.[4]

His efficiency argument favored letting models write programs that combine actions and filter results before returning them to the model. As the foundation’s account explained, this avoids repeated model turns for control flow that ordinary code, such as a loop, can handle. Cooper also anticipated higher-level delegation as services become AI-capable, while identifying discovery and verification of counterpart identities across providers, clients and hosts as an unresolved problem requiring community input.[4]