Stateless MCP 2.0: Revolutionary Update for AI Agents
Anthropic's Model Context Protocol offers simpler and safer agent frameworks.
1Launch of MCP 2.0
The Model Context Protocol (MCP) specification version 2026-07-28 has been released. This update is considered the most significant change to the protocol since its initial launch in 2024. Called Stateless MCP, the new version standardizes how AI agents interact with tools. Developed by Anthropic, the protocol gained significant attention in 2025 but was later overshadowed by another technology called Skills.
2From Stateful to Stateless
The old stateful MCP version required two HTTP requests: one to initialize a session and another to call a tool. The new stateless MCP works with a single HTTP request. This change significantly reduces implementation complexity on both client and server sides. It offers important scalability advantages for web applications by eliminating the need to maintain server-side session state.
3Security and Accessibility Benefits
Stateless MCP provides a more secure environment for AI agents. It reduces the risks associated with agents having shell environment and internet access. It also enables even smaller models to use this protocol effectively. MCP tools are preferred in security-focused scenarios as they're easier to audit and control. These features allow the protocol to be used effectively by models running on laptops.
4MCP Explorer Tool
A Python-based CLI tool called mcp-explorer was developed to test the protocol. This tool works with the uvx package manager and requires no installation. It allows users to query MCP servers, inspect tools, and make calls. For example, it can list tools on the agentic-mermaid.dev demo MCP or display the input/output schema of a specific tool. The tool returns results in JSON format, which can be processed with tools like jq.
5Datasette Integration
Thanks to the stateless MCP specification, an MCP plugin for Datasette was successfully developed. Called datasette-mcp, this plugin adds a /-/mcp endpoint to any Datasette instance. The plugin currently offers three basic tools: listing databases, getting database schema, and executing SQL queries (read-only for now). These tools can be connected to chat tools like ChatGPT or Claude, enabling users to send SQL queries to Datasette instances.
6LLM Integration
The new llm-mcp-client plugin provides official MCP integration for the LLM tool. This plugin allows users to access MCP servers directly from the LLM interface. The alpha version of the plugin enables users to MCP(