What the Agentic AI Foundation (AAIF) Means for MCP and the Future of Agentic AI
On December 9, 2025, Anthropic, Block, and OpenAI donated their most strategic AI agent projects to a neutral open foundation. Here's why the AAIF matters for everyone building with AI agents.
Mohammed Kafeel
Machine Learning Researcher
On this page
In under 13 months, the Model Context Protocol went from an internal Anthropic tool to the connective tissue of the entire AI agent ecosystem - adopted by Claude, ChatGPT, GitHub Copilot, VS Code, Cursor, and more than 10,000 published servers. That kind of velocity doesn't happen without a governance problem waiting to catch up with it.
On December 9, 2025, the Linux Foundation announced the Agentic AI Foundation (AAIF) - a neutral, open foundation purpose-built to make sure agentic AI grows in the public interest, not behind closed doors.
This isn't just another consortium press release. Three of the biggest names in AI - Anthropic, Block, and OpenAI - each donated a flagship project. Google, Microsoft, AWS, Cloudflare, and Bloomberg signed on as supporting founders. And by June 2026, a fourth project had joined.
Here's everything you need to know.
What Is the Agentic AI Foundation (AAIF)?
The AAIF is a directed fund under the Linux Foundation, launched on December 9, 2025, in San Francisco. Its job is to provide neutral stewardship for the open standards and tools that agentic AI runs on.
Think of it the way you'd think of the Linux Foundation itself - a neutral home where competing companies can collaborate on shared infrastructure without any single vendor calling the shots.
The official mission: ensure agentic AI evolves transparently, collaboratively, and in the public interest. Concretely, that means:
- Preventing vendor lock-in - no single company owns the protocols your agents depend on
- Eliminating fragmentation - one ecosystem, not ten incompatible ones
- Enabling AI agent interoperability - tools, models, and platforms that actually work together
You can explore the foundation directly at aaif.io. (For how MCP got here in the first place, see our MCP adoption timeline.)
Who's Behind It? The Founding Members
The AAIF didn't launch with a vague promise of future participation. It launched with committed code and committed companies.
Co-Founders (Project Donors)
Three organizations each donated a major project at launch:
- Anthropic → donated the Model Context Protocol (MCP)
- Block (Square, Cash App, Afterpay) → donated goose
- OpenAI → donated AGENTS.md
These aren't peripheral side projects. MCP is Anthropic's most widely adopted open-source contribution. goose is Block's flagship agentic framework. AGENTS.md is already used by more than 60,000 open-source projects.
Supporting Founding Members (Platinum)
The eight Platinum members - who fund the foundation and shape its direction - are:
- Amazon Web Services (AWS)
- Anthropic
- Block
- Bloomberg
- Cloudflare
- Microsoft
- OpenAI
That's a remarkable lineup. Competitors sitting at the same governance table, agreeing that the infrastructure layer should stay open.
Jim Zemlin, Executive Director of the Linux Foundation: "We are seeing AI enter a new phase, as conversational systems shift to autonomous agents that can work together. Within just one year, MCP, AGENTS.md and goose have become essential tools for developers building this new class of agentic technologies. Bringing these projects together under the AAIF ensures they can grow with the transparency and stability that only open governance provides."
The Four Core Projects
At launch, the AAIF housed three projects. In June 2026, a fourth joined. Here's how they compare:
| Project | Donated By | What It Does | Why It Matters |
|---|---|---|---|
| MCP (Model Context Protocol) | Anthropic | Universal open standard for connecting AI models to external data, tools, and APIs | The "HTTP for AI agents" - makes every integration reusable instead of custom-built |
| goose | Block | Open-source, local-first AI agent framework with MCP-based integration | Gives developers a structured, trusted runtime for building agentic workflows |
| AGENTS.md | OpenAI | Markdown standard that gives AI coding agents project-specific guidance | Makes agent behavior predictable across 60,000+ repos and toolchains |
| agentgateway | Solo.io | AI-native proxy and unified gateway for MCP, A2A, LLM inference, and API traffic | The infrastructure layer - routes, secures, and observes all agent traffic in one place |
A Closer Look at agentgateway
agentgateway joined the AAIF in June 2026, contributed by Solo.io, making it the foundation's fourth hosted project. It's an AI-native proxy - think of it as the gateway that sits in front of all your agent traffic.
It handles:
- MCP and Agent-to-Agent (A2A) routing for agent interoperability protocols
- LLM inference traffic with model-independence (switch providers without rewriting code)
- REST and gRPC API traffic through a single unified control plane
- Security controls - JWT auth, RBAC, mTLS, prompt guards, and more
- Built-in observability - metrics, tracing, and access logs for AI workflows
Licensed under Apache 2.0, agentgateway has over 300 active contributors across 60+ organizations, including CoreWeave, Red Hat, Adobe, Salesforce, and Microsoft. You can read the full announcement at agentgateway.dev.
Why MCP Is the Star of the Show
MCP - the Model Context Protocol - is the reason most developers first heard about the AAIF. And for good reason. (New to the protocol? Start with what MCP is.)
The Problem MCP Solved
Before MCP, connecting an AI model to an external tool meant writing a custom integration. Every time. For every tool. For every model.
If you had 10 AI tools and 10 data sources, you potentially needed 100 custom connectors. That's the classic M×N integration problem - fragmented, expensive, and nearly impossible to scale.
MCP solved it with a single universal standard. One protocol. One integration pattern. Write it once, connect to everything.
Mike Krieger, Chief Product Officer at Anthropic, put it plainly: "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."
What MCP Actually Does
MCP standardizes how AI models:
- Request files from local or remote storage
- Query databases without custom connectors
- Call external APIs through a consistent interface
- Execute actions in tools like GitHub, Slack, or Notion
The architecture is a clean client-server model:
- Hosts - the AI application (e.g., Claude Desktop, VS Code)
- Clients - the AI agent making requests
- Servers - external services exposing data or tools via MCP
The three core primitives are Tools (functions the AI can call), Resources (data the AI can read), and Prompts (pre-defined workflow templates).
MCP's Adoption Numbers
The scale here is hard to overstate:
- 10,000+ published MCP servers as of December 2025
- Adopted by Claude, ChatGPT, GitHub Copilot, Cursor, VS Code, Gemini, and more
- Deployed at enterprise scale on AWS, Google Cloud, and Azure
Anthropic often describes MCP as "USB-C for AI" - just as USB-C standardized how hardware connects, MCP standardizes how AI software connects to the world. (Ready to build one? See our guide to building an MCP server in Python.)
How Is AAIF Governed?
Open governance is only as good as its structure. The AAIF's is worth understanding - especially if you're building on any of these projects.
Governing Board
David Nalley, Director of Developer Experience at AWS, was appointed Governing Board Chair in February 2026. The board is responsible for the foundation's strategic direction, budget, and membership.
Technical Committee
The Technical Committee has one representative from each of the eight Platinum members. It:
- Defines technical requirements and standards
- Reviews project proposals for inclusion
- Manages the project lifecycle: Growth → Impact → Emeritus stages
- Holds biweekly open meetings on the LFX platform - anyone can attend
Seven Working Groups (Launched Q1 2026)
The AAIF stood up seven working groups in Q1 2026, each tackling a specific challenge in the agentic AI landscape:
- Identity and Trust - how agents authenticate and verify each other
- Accuracy and Reliability - reducing hallucinations and improving agent dependability
- Workflows and Process Integration - connecting agents to business processes
- Agentic Commerce - enabling agents to transact on behalf of users
- Security and Privacy - protecting data in multi-agent systems
- Observability and Traceability - logging and monitoring agent actions
- Governance Risk and Regulatory Alignment - keeping agentic AI compliant with emerging regulations
These aren't just committees on paper. They're where the real technical work happens - and they're open to community participation.
What Does This Mean for the Future of Agentic AI?
Here's the honest answer: the AAIF is the most significant structural development in agentic AI since MCP itself.
Open Standards Win
History is consistent on this. HTTP beat proprietary web protocols. Linux beat proprietary Unix. Kubernetes beat proprietary container orchestration. When a neutral foundation governs critical infrastructure, adoption accelerates and the ecosystem expands.
MCP under AAIF governance means you can build on it without worrying that Anthropic will pivot, monetize, or restrict access. That's not a small thing. (Open governance is already shaping the protocol's direction - see the MCP 2026 roadmap.)
Interoperability Becomes Real
The combination of MCP (protocol), goose (runtime), AGENTS.md (instructions), and agentgateway (traffic control) gives the ecosystem a full stack for AI agent interoperability. These aren't four random projects - they're four layers of the same infrastructure.
An agent built with goose, guided by AGENTS.md, connecting via MCP, and routed through agentgateway can work with any compliant tool, model, or platform. That's the promise of open standards made concrete. (For how MCP relates to the agent-to-agent layer, see MCP vs A2A.)
What Developers Should Expect
If you're building with agentic AI frameworks today, here's what the AAIF changes for you:
- Stability: Projects under Linux Foundation governance don't disappear or get acquired into a closed product
- Community input: The biweekly Technical Committee meetings and working groups are open - you can participate
- Faster ecosystem growth: Neutral governance lowers the barrier for other companies to build MCP servers, goose extensions, and agentgateway plugins
- Regulatory readiness: The Governance Risk and Regulatory Alignment working group is already working on compliance frameworks before regulators force the issue
The agentic AI future isn't just about smarter models. It's about the infrastructure those models run on. The AAIF is building that infrastructure in the open.
Key Takeaways
TL;DR - The most important things to remember:
- 🏛️ The AAIF launched December 9, 2025, under the Linux Foundation, with Anthropic, Block, and OpenAI as co-founders
- 🔌 MCP is now community-governed - no single vendor controls the protocol your agents depend on
- 🧰 Four projects, one ecosystem: MCP (protocol), goose (runtime), AGENTS.md (instructions), agentgateway (gateway)
- 🏢 180+ member organizations by mid-2026, including AWS, Google, Microsoft, Bloomberg, and Cloudflare
- 🗳️ Seven working groups are actively shaping standards for identity, security, observability, commerce, and more
- 🚀 For developers: build on these projects with confidence - they're stable, open, and community-driven
- 📅 David Nalley (AWS) chairs the Governing Board; biweekly Technical Committee meetings are open to all
FAQ
What is the Agentic AI Foundation (AAIF)?
The AAIF is a directed fund under the Linux Foundation, launched on December 9, 2025, in San Francisco. It provides neutral, open governance for the key standards and tools powering agentic AI - including MCP, goose, AGENTS.md, and agentgateway. Its mission is to ensure agentic AI evolves transparently and in the public interest.
What is the Model Context Protocol (MCP)?
MCP is an open standard that lets AI models connect to external data sources, tools, and systems through a single universal protocol. Instead of building custom integrations for every tool, developers implement MCP once and gain access to thousands of compatible servers. It's often described as "HTTP for AI agents" or "USB-C for AI."
Who founded the AAIF?
The AAIF was co-founded by Anthropic, Block, and OpenAI, who each donated a flagship project. Supporting Platinum members include AWS, Bloomberg, Cloudflare, Google, and Microsoft. The Linux Foundation serves as the neutral governing body.
What is agentgateway, and why did it join the AAIF?
Agentgateway is an open-source, AI-native proxy contributed by Solo.io. It joined the AAIF in June 2026 as the fourth hosted project. It provides a unified gateway for MCP traffic, Agent-to-Agent (A2A) communication, LLM inference, and conventional API traffic - all under one security and observability layer. It's Apache 2.0 licensed with 300+ contributors across 60+ organizations.
How is the AAIF governed?
The AAIF has a Governing Board (chaired by David Nalley of AWS, appointed February 2026), a Technical Committee with one rep from each of the eight Platinum members, and seven working groups covering identity, security, observability, commerce, workflows, accuracy, and regulatory alignment. Technical Committee meetings are biweekly and open to the public via the LFX platform.
What is goose?
Goose is an open-source, local-first AI agent framework donated by Block. It combines language models, extensible tools, and MCP-based integration to give developers a structured, reliable environment for building and running agentic workflows. It was released in early 2025 and is now governed by the AAIF.
What is AGENTS.md?
AGENTS.md is a markdown-based standard released by OpenAI in August 2025 and donated to the AAIF. It gives AI coding agents a consistent source of project-specific guidance - things like coding conventions, repo structure, and workflow rules - so agent behavior is predictable across different repositories and toolchains. It's already adopted by 60,000+ open-source projects.
Is the AAIF open to new members and projects?
Yes. The AAIF accepts new member organizations at Platinum, Gold, and Silver tiers. It also has an open project intake process managed by the Technical Committee. Projects go through Growth, Impact, and Emeritus lifecycle stages. Details are at aaif.io.
Useful Sources
Keep reading
MCP Adoption Timeline: From Anthropic Experiment to Linux Foundation Standard
A complete history of the Model Context Protocol (MCP) timeline: from launch to Linux Foundation donation, SDK downloads, and enterprise adoption.
Human-in-the-Loop MCP Workflows: When Agents Should Pause for Approval
A practical guide to adding human approval checkpoints to MCP-powered AI agent workflows — MCP elicitation, approval patterns, a 5-trigger decision framework, and real-world use cases.
The Three-Layer AI Agent Stack: MCP, A2A, and Streamable HTTP Explained
MCP, A2A, and Streamable HTTP are the three protocols that form the modern AI agent stack. Here's exactly how they fit together — and why it matters for every developer building with AI.