The brand-new Agentic AI Foundation has been launched by the Linux Foundation. Three projects form the basis for this new foundation: Anthropic’s hugely successful Model Context Protocol (MCP), Block’s goose, and OpenAI’s AGENTS.md.
Of the three contributions, MCP is undoubtedly the best known. In November 2024, Anthropic launched the protocol on an open-source basis, after which it became commonplace within the AI community within a few months. It is the standard protocol that allows AI models to perform actions via tools, based on their own data and in communication with external applications. In short, it is a cornerstone for agentic AI. For this reason, it is not surprising that the foundation established by the Linux Foundation is called the Agentic AI Foundation (AAIF). A month ago, MCP received a major update from its creator, Anthropic, primarily to raise the fairly basic security level and add much-requested features such as support for complex workflows.
Read more about MCP: How the Model Context Protocol has taken the AI world by storm
goose and AGENTS.md
The other two projects that the AAIF is working on are goose from Block and AGENTS.md from OpenAI. Block is the company behind payment platforms Square, Cash App, Afterpay, and TIDAL. It developed goose as an AI agent framework and integrated MCP into it from the start. The goal behind the implementation was to provide a stable local environment for agentic workflows. In other words, it should make AI rollout practical for organizations by ensuring security and consistency. The combination of MCP and goose is therefore not just logical, it is a requirement for the further development of agentic AI to ever be called “enterprise-ready.”
AGENTS.md also aims to make AI behavior more deterministic in software environments. It is a standard for AI coding agents so that they continuously perform the correct actions to consult or manipulate repositories and chains of tools.
Together, the three donated projects should build a path towards workable agentic AI for the AAIF. It is a big step for MCP to be transferred to this foundation, because this truly guarantees neutrality despite all of Anthropic’s commitments. As a contributor, Claude’s creator will undoubtedly remain active, but as part of the Linux Foundation, the AAIF is a combination of breakthroughs from multiple parties. First and foremost, they must remain in harmony as they become more powerful.
AI in check
We know that MCP actually originated as an internal Anthropic tool. It was designed to solve the problem that there was no consistent way for AI to communicate with and act on the outside world. Ad hoc solutions have now been phased out now that MCP is universal. The fact that it has spread so quickly shows that other AI developers were eager for such a solution. However, as is often the case with highly successful open-source contributions, MCP is not a protocol that prioritizes security. That is not necessary either. Kubernetes, for example, is also a technology that you still have to secure yourself, but it helps to get a coordinated, secure-by-default solution.
Everyone agrees that this should be done on an open-source basis. It prevents duplication of effort for AI players and integration problems for users of LLMs and agentic AI systems. We wonder whether Agent2Agent (A2A), a Google protocol for agents to work together, can also become part of the AAIF or be developed to achieve similar best practices as MCP, goose, and AGENTS.md. It has already been donated to the Linux Foundation, so the step to the AAIF is not too big.
Read also: Microsoft Agent Framework: multi-agent systems take shape