What Modelcontextprotocol changed

• Replaced the general "What is MCP?" introduction with a formal "Specification" document.

• Added a detailed "Security and Trust & Safety" section outlining principles for user consent and data privacy.

• Defined specific roles for the protocol: Hosts (LLM apps), Clients (connectors), and Servers (data providers).

• Introduced technical requirements for "Sampling," "Roots," and "Elicitation" features.

• Updated the versioning from a future-dated placeholder (2026) to a specific release (2025-11-25).

• Added a "Key Changes" changelog and a formal "Schema Reference."

• Replaced the "What is MCP?" introduction with a deep-dive "Architecture overview" page.

• Added detailed technical sections for the Data Layer and Transport Layer.

• Introduced a step-by-step walkthrough of client-server interactions using JSON-RPC 2.0 messages.

• Defined specific roles for "MCP Hosts" (like Visual Studio Code) versus "MCP Clients."

• Formally deprecated "Sampling" and "Logging" primitives in favor of direct LLM API calls and OpenTelemetry.

• Added a new "Elicitation" primitive that allows servers to ask users for more information.

• Added new feature flags for "turbopuffer-serve," "live-agents," and "comment-mode."

• Updated the internal client version from 0.0.4067 to 0.0.4074.

• Modified internal CSS and configuration metadata for documentation rendering.

The project has moved from general documentation to a formal technical specification. This provides a stable foundation for developers to build interoperable AI tools and defines strict rules for how "Hosts," "Clients," and "Servers" must communicate.

The protocol is maturing with clearer definitions of how AI apps (hosts) connect to data (servers). The new documentation provides a technical roadmap for developers to build more complex integrations using JSON-RPC 2.0.

The platform is enabling new backend capabilities for "live agents" and "comment mode." This suggests the protocol is expanding to support real-time, interactive AI agent workflows rather than just static data retrieval.

• Updated the version from 2026-07-28 to 2025-11-25.

• Replaced the general introduction with a formal technical specification document.

• Added a detailed "Security and Trust & Safety" section covering user consent and data privacy.

• Introduced specific definitions for "Sampling," "Roots," and "Elicitation" as client features.

• Changed navigation from a "Get Started" focus to a "Specification" focus.

• Added references to official IETF standards (RFC 2119) for protocol requirements.

• Replaced the basic "What is MCP?" introduction with a deep-dive "Architecture Overview."

• Added detailed technical sections on the Data Layer (JSON-RPC 2.0) and Transport Layer (Stdio vs. HTTP).

• Introduced a step-by-step walkthrough of client-server interactions, including discovery and tool execution.

• Formally deprecated "Sampling" and "Logging" primitives in favor of direct LLM API integration and OpenTelemetry.

• Clarified the roles of "MCP Host" (e.g., Claude Desktop), "MCP Client," and "MCP Server."

• Added detailed CSS for a new "schema-reference" section to improve the display of technical specifications.

• Updated font weights for heading levels (h5 and h6) to improve visual hierarchy.

• Added a custom counter system for section numbering in the documentation.

• Improved code block styling, including specific colors for line highlights and JSON syntax.

• Updated the client version from 0.0.4054 to 0.0.4067.

The project has shifted from a general "Get Started" guide to a formal technical specification. This signals the protocol is maturing and moving toward a standardized, stable version for developers to follow.

The protocol is maturing with clearer definitions of how AI applications (hosts) connect to data sources (servers). This update provides the technical blueprint for developers to build more complex, multi-server AI tools.

The site updated its styling and technical documentation layout. This improves how developers read and navigate their code standards and schema references.

• Replaced the general "What is MCP?" introduction with a formal technical specification.

• Added a detailed "Security and Trust & Safety" section covering user consent and data privacy.

• Defined specific "Client Features" including Roots, Sampling, and Elicitation.

• Updated the protocol versioning from a future-dated 2026-07-28 to 2025-11-25.

• Introduced formal "MUST/SHOULD" requirements based on IETF standards (RFC 2119).

• Replaced the basic "Getting Started" landing page with a comprehensive "Architecture Overview."

• Added technical details for the Data Layer (JSON-RPC 2.0) and Transport Layer (Stdio and HTTP).

• Introduced a step-by-step walkthrough of client-server discovery and tool execution.

• Defined core primitives: Tools, Resources, and Prompts.

• Marked "Sampling" and "Logging" as deprecated features in the latest protocol version.

• Added specific examples of host applications like Visual Studio Code and Claude Desktop.

• Added a new feature toggle for "rbac-expansion" in the site configuration.

• Updated the internal client version from 0.0.4045 to 0.0.4054.

• Modified internal documentation metadata related to security best practices.

The project is moving from general documentation to a formal technical standard. This makes it easier for developers to build reliable tools that work together using a strict set of rules.

The project has moved from a simple introduction to a technical deep dive. This helps developers understand exactly how the data and transport layers work, making it easier for them to build complex integrations that compete with your current offerings.

The addition of Role-Based Access Control (RBAC) suggests the protocol is maturing to handle complex enterprise security needs and permission management.