level: technical
The Model Context Protocol (MCP) 2.0 specification, released on July 28, 2026, introduces a stateless design that replaces session-based interactions with single HTTP requests. This change reduces complexity for both clients and servers, eliminating the need for session IDs and state management. Simon Willison, who had lost interest in MCP due to the rise of more flexible agent approaches, now sees renewed value in the protocol for building auditable and controllable tools.
The stateless approach requires only one HTTP request to call a tool, compared to the legacy two-step process. This makes it easier to build scalable web applications without server-side state. Willison built three projects to explore the new spec: mcp-explorer, a CLI tool for probing MCP servers; datasette-mcp, a Datasette plugin exposing a read-only SQL interface; and llm-mcp-client, an alpha plugin for his LLM tool. These demonstrate practical integration with services like ChatGPT and Claude.
MCP offers a safer alternative to giving agents arbitrary shell and internet access, which poses significant security risks. By constraining tools to predefined interfaces, MCP makes it easier to audit and reason about agent capabilities. Willison plans to use MCP more in sensitive applications, noting that even smaller models can drive MCP tools effectively. The new spec has lowered the barrier for implementing both clients and servers, potentially broadening adoption.
why it matters: Stateless MCP simplifies building secure, auditable tool integrations for AI agents, reducing implementation complexity and security risks compared to open-ended command execution.