Model Context Protocol

What it is

The Model Context Protocol (MCP) is an open standard that defines how an AI assistant connects to outside systems: file stores, messaging tools, code hosting, databases and similar sources of data. Once a system speaks MCP, any assistant that understands the protocol can use it, and the same system does not need a separate integration for each assistant.

Origins

Anthropic open-sourced MCP on 25 November 2024. Its reasoning was that every new data source needed its own custom integration. MCP was meant to replace these with one protocol.^[1]

The launch included the specification and software development kits (SDKs), support for local MCP servers in the Claude desktop apps, and open-source servers for Google Drive, Slack, GitHub, Git, Postgres and Puppeteer.^[1] Early adopters were Block and Apollo. The development-tool companies Zed, Replit, Codeium and Sourcegraph were also working with it.^[1]

In May 2025 Anthropic added an MCP connector to its API, so that Claude could call remote MCP servers directly.^[4]

How it works

MCP is a specification with versions. The versions are dated in the form YYYY-MM-DD, and a new version number is used only for backwards-incompatible revisions, so a date tells a developer which rules apply. The current version is dated 28 July 2026. It replaced the handshake-based revisions that came before, including the one dated 25 November 2025 and earlier.^[3]

The 25 November 2025 revision added asynchronous operations, statelessness, server identity and official extensions. A community MCP Registry, where servers can be listed and found, was published alongside it.^[2]

Put simply, an MCP server exposes capabilities such as reading a file or searching a database, and an MCP client, which is the assistant or the application around it, discovers and calls them through the protocol. A useful way to picture this is a standard plug: a device needs only one kind of socket to be compatible with many outlets.

Governance

Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded the fund, and Google, Microsoft, AWS, Cloudflare and Bloomberg support it.^[2] A directed fund is a body set up to take responsibility for an open project so that no single company controls it.

At the time of the donation, Anthropic reported more than 10,000 active public MCP servers and 97 million monthly SDK downloads. It also said that ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code had adopted the protocol.^[2] In October 2026 Anthropic also donated Agent Skills to the same foundation.^[5]

Why it matters

MCP matters for three reasons. It reduces the repeated work of building one integration for each assistant and each data source. It lets data stay where it already is, with access granted through a server rather than by copying data into a model's context. And it places the standard under a neutral foundation, so the protocol's future does not depend on one company.^[1][2]

MCP is complementary to Agent Skills, which provide reusable instructions, rather than a substitute for them. A skill can tell Claude how to do a task, while MCP provides the connection to the systems the task needs.^[5]