New Version of the MCP Specification (modelcontextprotocol.io)

The Model Context Protocol (MCP) specification has been updated and published as the authoritative TypeScript schema (schema.ts) to standardize how LLM applications connect to external data sources and tools. MCP uses JSON-RPC 2.0 over stateful connections with explicit host/client/server roles and capability negotiation, and is inspired by the Language Server Protocol. Servers can expose Resources (context/data), Prompts (templated workflows) and Tools (callable functions); clients can enable Sampling (server-initiated agentic behaviors and recursive LLM calls), Roots (server-initiated filesystem/URI scopes) and Elicitation (requests for additional user input). The spec also defines utilities for configuration, progress tracking, cancellation, error reporting, logging and security metadata, and points to modelcontextprotocol.io for implementation guides. This update is significant because it provides a common, extensible contract for composing LLM-driven workflows, enabling richer tool use, model-initiated actions, and interoperable integrations across IDEs, chat interfaces and custom apps. At the same time MCP explicitly addresses risk: implementors must build consent/authorization flows, preserve user control over data and tool invocation, limit server visibility into prompts, and treat tool annotations as untrusted unless verified. The spec emphasizes privacy, access controls and explicit user approval for LLM sampling and tool execution—practical guardrails that balance powerful agentic capabilities with safety and deployability.