Anthropic, OpenAI, and Block donate AI agent projects to new Linux Foundation body

The Linux Foundation today announce the formation of the Agentic AI Foundation (AAIF), a new home for open source projects that underpin how AI agents connect to tools, data, and each other. The technology matters, but so does who's contributing it.

Anthropic is donating the Model Context Protocol (MCP). OpenAI is contributing AGENTS.md. Block is contributing Goose. These are direct competitors, and they've each donated core projects to neutral governance.

The membership list reinforces the point - Amazon, Google, Microsoft, Bloomberg, and Cloudflare are Platinum members. Gold members include Cisco, IBM, Oracle, SAP, and Snowflake.

What these projects actually do

If you're building applications that use AI agents - software that can take actions autonomously rather than just answering questions - you face a practical problem: how does the agent connect to your existing tools and data?

MCP, or Model Context Protocol, is Anthropic's answer. It provides a standardized way for AI models to connect to external tools, databases, and applications. MCP acts as a universal adapter. Rather than building custom integrations for every combination of AI model and business tool, MCP provides a common protocol. It launched in November 2024 and has been adopted rapidly - there are now more than 10,000 published MCP servers, and the protocol is supported by Claude, ChatGPT, Microsoft Copilot, Gemini, and most major coding tools.

Mike Krieger, Chief Product Officer at Anthropic, explains:

MCP started as an internal project to solve a problem our own teams were facing. When we open sourced it in November 2024, we hoped other developers would find it as useful as we did. A year later, it's become the industry standard for connecting AI systems to data and tools.

Goose, from Block (the company behind Square and Cash App), is an AI agent framework. Where MCP handles connections, Goose provides the structure for building agents that can actually do things - writing code, running tests, managing files, executing workflows. It runs locally on your machine and can work with any Large Language Model (LLM) that supports tool calling. Goose uses MCP for its integrations, which is partly why Block and Anthropic ending up in the same foundation makes sense.

AGENTS.md is simpler but solves a real problem. When an AI coding agent works on a project, it needs to understand project-specific conventions - how to run tests, what build system to use, which files to avoid touching. AGENTS.md is a markdown file format that provides this guidance in a standardized way. OpenAI released it in August 2025, and it has already been adopted by more than 60,000 open source projects.

Earlier signals

At Open Source Summit EU in Amsterdam in August, Linux Foundation executives and engineers from Amazon Web Services (AWS), Cisco, and others were already discussing this trajectory. Mark Collier, General Manager of AI Infrastructure at the Linux Foundation, talked about bringing agent-related projects under neutral governance. The conversation kept returning to the same themes: trust, identity, interoperability, and the need for standards that no single company controls.

One participant made the comparison to containers - that standardization process took roughly a decade, from early container technologies through Docker to Kubernetes becoming the de facto orchestration platform. The agent ecosystem is moving faster, but faces similar challenges. As systems become more autonomous, the questions of identity (which agent is this?), observability (what is it doing?), and security (should it be allowed to do that?) become more pressing.

The AAIF doesn't solve these problems yet. But it provides a neutral space where they can be worked on collaboratively.

What this means in practice

If you're evaluating agentic AI tooling for your organization, this changes the picture. Building on MCP becomes safer when it's governed by a foundation rather than a single vendor. The same applies to the other contributed projects.

Shawn Edwards, Chief Technology Officer at Bloomberg, frames it in terms of financial services requirements:

MCP provides the essential connective layer required in our work building and deploying agentic AI systems for finance that do far more than simple question-answering. As an open source standard governed by the Linux Foundation, MCP is poised to drive broader adoption and innovation across the financial sector.

Each Platinum member can appoint a representative to AAIF's Governing Board - Amazon, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, and OpenAI all have seats.

My take

At the same Open Source Summit in Amsterdam, Dirk Hohndel, Head of Open Source Program Office at Verizon, gave a keynote that stuck with me. He pointed out that 95% of the software stacks we use every day are open source, but 99% of corporate spending goes to the remaining 5% that's proprietary. Companies will spend hundreds of millions on proprietary software while struggling to justify $10,000 to support an open source project.

He asked the audience to think about what it means to control your own digital infrastructure - your data, your tools, your ability to operate independently. The alternative is being locked into systems controlled by companies whose interests may not align with yours.

The AAIF represents companies choosing the open path for this particular layer of technology. Anthropic, OpenAI, and Block could have kept MCP, AGENTS.md, and Goose proprietary. They decided not to. Whatever mix of principle and pragmatism drove that decision, the outcome is the same: critical infrastructure moving to neutral ground.

The hard problems remain unsolved. Agent identity, security, observability, and governance are all still works in progress. But the foundation is now in place - literally - for that work to happen collaboratively rather than in competing silos.