Model Context Protocol (MCP) Specification 2025-11-25 (mcp.mintlify.app)

The Model Context Protocol (MCP) specification (2025-11-25) formalizes an open, JSON-RPC‑based protocol for connecting LLM applications to external data sources, tools and workflows. Inspired by the Language Server Protocol, MCP defines roles (Hosts, Clients, Servers), stateful JSON-RPC 2.0 messaging, and a TypeScript schema (schema.ts) as the authoritative contract. It’s designed to make integrations composable and interoperable across IDEs, chat interfaces and custom AI workflows—reducing fragmented ad‑hoc connectors and enabling a common ecosystem for exposing contextual resources, templated prompts and executable tools to models. Technically, MCP standardizes server-offered features (Resources, Prompts, Tools) and client-offered behaviors (Sampling, Roots, Elicitation), plus utilities for configuration, progress, cancellation, error reporting and logging. Crucially, the spec embeds security and trust principles: explicit user consent for data access and tool invocation, strict data‑privacy constraints, limited server visibility into prompts, and explicit controls around LLM sampling and recursive agentic behavior. While the protocol cannot enforce UI or policy, implementors are urged to build robust consent/authorization flows, access controls and threat mitigations. The spec and implementation examples are available at modelcontextprotocol.io, offering a blueprint for safer, standardized tooling and data integration in the LLM ecosystem.