Model Context Protocol turns one, releases new spec version (blog.modelcontextprotocol.io)

The Model Context Protocol (MCP) celebrates its first anniversary and has released a new specification version. What began as an open‑source experiment to provide structured context to LLMs has become the de‑facto integration standard for connecting data, tools, and applications to models. Adoption exploded in under a year: active MCP servers grew from a handful to thousands, the MCP Registry now catalogs nearly 2,000 entries (a 407% increase since September), and major platforms—GitHub, OpenAI, Google, Microsoft, AWS, Hugging Face, Stripe, Notion and others—have shipped servers, SDKs and integrations that surface data, workflows and model management through MCP. Technically, MCP standardizes how agents and LLMs discover, request and receive external context, enabling write‑once, integrate‑everywhere interoperability for agentic systems. The community has driven Specification Enhancement Proposals, SDKs, transports, security extensions and tooling (e.g., MCP Inspector), and formalized governance (SEP‑1302 and working groups) to evolve the spec without breaking implementations. Recent ecosystem work emphasizes authorization (Cross App Access), human‑in‑the‑loop patterns, asynchronous execution, and richer model/dataset discovery—reducing fragmentation and making agent pipelines more secure, auditable and portable across clouds and applications. The new spec release, backed by a broad contributor base and enterprise adopters, signals MCP maturing from experiment to foundational infrastructure for production AI.