Anthropic donates MCP to the new Agentic AI Foundation

On 9 December 2025, Anthropic donated the Model Context Protocol (MCP) to the Agentic AI Foundation (AAIF), a new directed fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded the fund, with support from Google, Microsoft, AWS, Cloudflare and Bloomberg. MCP sits alongside two other founding projects, Block's goose and OpenAI's AGENTS.md.

What MCP does

Anthropic introduced MCP in November 2024 as an open standard for connecting AI systems to the business software they need to read from and act on. Anthropic compared it to a USB-C port for AI applications, and the comparison holds, because before a shared port every device needed its own cable. Before MCP, connecting an AI tool to your CRM, document store or ticketing system meant a custom integration, built for one pair of systems and repaired whenever either side changed. With MCP, a system exposes its capabilities once in a standard way, and any compliant AI tool can connect to it.

What changed with the donation

According to Anthropic and the Linux Foundation, MCP is now governed by a vendor-neutral foundation instead of the company that created it. Companies that compete on AI models, including OpenAI, Google, Microsoft and AWS, now support the same body that looks after the protocol. On 21 July 2026 the AAIF went a step further and launched the Model Context Protocol Associate (MCPA), which it describes as the first official certification for MCP. The protocol now has an independent home, a group of backers that includes most of the large AI and cloud providers, and a recognised credential.

What it means for engineering teams

Your integrations become less tied to a single vendor. When your AI layer talks to your systems over a standard that no single AI company controls, replacing the model or the tool no longer means rebuilding the connections underneath. You also gain a simple question for every AI vendor meeting, which is whether the product speaks MCP and, if it doesn't, what you'd be paying to build instead.

Your team takes on a new kind of risk at the same time. Every MCP server is code that can read data and take actions inside your systems, so it deserves the same security review as any other third-party dependency. Engineers should check who maintains a server and which permissions it requests before it goes anywhere near production, and every action an AI tool takes through it should be logged.

What it means when you hire engineers

Building and reviewing MCP servers is now a skill you can name in a job brief, and the MCPA gives you an outside reference point for it. A certification shows that an engineer has learned the protocol. It can't show how they scope access in a messy production codebase or how they test what an AI agent can and can't touch. Engineers who've built MCP servers should be able to walk you through those decisions on a real system.

In our vetting, every candidate does two sixty-minute assessments in a real codebase, the first with AI tools off and the second with the candidate's own agents on. You get both scorecards, so you can see how an engineer works with these tools as well as without them.

You can read more about how we vet engineers, including what each session covers.