Anthropic, Block and OpenAI handed MCP, goose and AGENTS.md to a neutral Linux Foundation fund. Here is what AAIF actually governs — and whether your MCP code changes.
What is the Agentic AI Foundation (AAIF)?
The Agentic AI Foundation (AAIF) is a neutral, vendor-independent home under the Linux Foundation that now stewards three of the most widely used open standards for AI agents: Anthropic’s Model Context Protocol (MCP), Block’s goose agent framework, and OpenAI’s AGENTS.md. It launched on December 9, 2025 as a directed fund inside the Linux Foundation, co-founded by Anthropic, Block and OpenAI, with backing from the largest cloud and AI vendors.
The agentic AI foundation exists to do one thing the original single-vendor projects could not: guarantee that the plumbing of the agent era evolves transparently and cannot be unilaterally controlled, monetized, or deprecated by the company that invented it. When Anthropic owned the MCP spec, every enterprise betting on it was, on paper, betting on Anthropic’s continued goodwill. Moving it into a Linux Foundation directed fund converts that bet into something a procurement and legal team will actually sign.
Every other article on this topic stops at the announcement: “Company X donated Project Y.” This guide does the part they skip — it translates governance neutrality into builder consequences. Who controls the MCP roadmap now? Can AWS or Google steer the spec because they pay for a Platinum seat? And the question every engineer is actually Googling: does my MCP server code change? (Short answer, covered in depth below: no.)
If you are new to the underlying standards, alatirok’s primers on what MCP is, the A2A protocol, and Agent Skills give the protocol-level background; this piece is about who governs them and what that means for the code you ship.
What does the Agentic AI Foundation govern (MCP, goose, AGENTS.md)?
AAIF governs four hosted projects: the Model Context Protocol (MCP) for connecting agents to tools and data, goose for running local-first agents, AGENTS.md for giving coding agents per-repository instructions, and agentgateway, a unified policy and routing layer for MCP and agent traffic. The first three were the founding donations; agentgateway joined as the foundation’s enterprise-facing infrastructure project.
These projects sit at different layers of the agent stack, which is exactly why neutral stewardship matters — together they describe how an agent finds tools, how it runs, how it reads project context, and how an enterprise governs the traffic in between. AGENTS.md alone has been adopted by more than 60,000 open source projects and is read by Codex, Cursor, Devin, Factory, Gemini CLI, GitHub Copilot, Jules, Amp and VS Code, per the Linux Foundation. goose is Block’s extensible, local-first agent that speaks MCP natively.
The table below maps each project to its origin, layer, and what changed for users at donation. The headline for all of them is identical: the home moved, the format did not.
AAIF is stewardship, not a rewrite. MCP, goose and AGENTS.md kept their specs, formats and maintainers. What changed is the legal and governance wrapper around them — they now live in a neutral Linux Foundation fund instead of a single company’s GitHub org.
| Project | Donated by | What it does | Layer | Did the spec/format change? |
|---|---|---|---|---|
| Model Context Protocol (MCP) | Anthropic | Connects agents to tools, data and apps | Integration / wire protocol | No — stewardship only |
| goose | Block | Local-first, extensible AI agent runtime | Agent runtime | No — same framework, new home |
| AGENTS.md | OpenAI | Per-repo instructions for coding agents | Project context / config | No — same Markdown convention |
| agentgateway | AAIF (community) | Policy, routing and governance for MCP/agent traffic | Gateway / enforcement | New AAIF-hosted project |
Who owns MCP now, and does my MCP server code change?
No — your MCP server and client code does not change. Anthropic donated stewardship of the Model Context Protocol to the agentic AI foundation, but the specification itself did not fork, the wire protocol is unchanged, and the same maintainers still control the roadmap. If you wrote an MCP server in 2025, it speaks the same protocol in 2026. There is nothing to migrate, no namespace to rename, no version pin to chase.
Anthropic was explicit that “the Model Context Protocol’s governance model will remain unchanged: the project’s maintainers will continue to prioritize community input and transparent decision-making.” In practice that means lead core maintainer David Soria Parra and the existing maintainer group still set direction through Spec Enhancement Proposals (SEPs) and working groups — the same process used before the donation. The donation transferred the trademark and the institutional home, not commit rights to a new committee.
This is the precise point every press-release rewrite glosses over. “Anthropic donated MCP” sounds like a handover that could break your integration. It is the opposite: it is the legal guarantee that no single company — including Anthropic — can now break it for you. The protocol you build against is more stable after the donation, not less.
“Stewardship moved. The spec did not fork. If you wrote an MCP server in 2025, it speaks the exact same protocol in 2026 — there is nothing to migrate.”
alatirok analysis of the AAIF donation
Concretely, what should I check in my codebase?
Nothing breaks, but two hygiene items are worth a pass: (1) Update any docs or READMEs that describe MCP as ‘Anthropic’s protocol’ — it is now an AAIF/Linux Foundation project. (2) Track the modelcontextprotocol.io spec repo for SEPs the same way you did before; the roadmap process is unchanged but the velocity is higher under foundation governance. There is no code change, no dependency bump, and no transport change required by the donation itself.How is AAIF governed, and can Platinum vendors steer the MCP spec?
AAIF separates money from merge rights. A governing board — chaired by AWS — controls budget, membership and strategic investment, while each project keeps full technical autonomy: MCP’s maintainers, not the board, decide what goes into the spec. So a Platinum vendor cannot buy its way into the MCP roadmap. Paying for a Platinum seat funds the foundation and earns board influence over its direction and spending; it does not grant commit access or a vote on Spec Enhancement Proposals.
This two-track design is the heart of how a Linux Foundation directed fund differs from a company-run open source project. The Technical Committee and per-project maintainer groups own the technical decisions; the Governing Board owns the institutional ones. At the MCP Dev Summit North America in April 2026, the Technical Steering Committee approved a formal three-stage project lifecycle policy — Growth, Impact, and Emeritus — opening the door for outside projects to join for the first time. That is governance maturing on the technical side, independent of which vendors write the biggest checks.
There is one nuance builders should internalize: maintainers from Anthropic, AWS, Microsoft and OpenAI drew a deliberate scope line at the summit — MCP connects applications to data and tools, and it explicitly excludes identity, observability and governance from the core protocol. That is a feature, not a gap. It is also where your responsibility (and the agentgateway project) begins.
The mental model: a Platinum membership buys a seat at the budget table, not a key to the spec repo. AWS chairing the board does not let AWS dictate MCP; the maintainers still merge the SEPs.Who are the AAIF members (Platinum and Gold)?
AAIF’s eight Founding Platinum members are AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI — effectively every major frontier-model lab and hyperscaler in one fund. Below them, a deep Gold tier includes Okta, Datadog, Docker, IBM, JetBrains, Salesforce, SAP, Shopify, Snowflake, Temporal and Twilio, with Silver members spanning Hugging Face, Pydantic, WorkOS, Uber, SUSE and others.
The membership roster is itself the strongest evidence of neutrality. When OpenAI, Anthropic, Google, Microsoft and AWS all sit at the same Platinum tier of the agentic AI foundation, no single one of them can capture the standard — their mutual presence is the check. That is structurally different from a protocol controlled by one lab, and it is why MCP crossed from interesting-standard to default-infrastructure in under a year.
The table below lists the tiers as reported by the Linux Foundation, Anthropic, and AAIF’s member page.
Neutrality is not a slogan here — it is enforced by the guest list. With OpenAI, Anthropic, Google, Microsoft and AWS all at Platinum, the standard belongs to the commons because no single member can dominate it.
| Tier | Named members |
|---|---|
| Platinum (founding) | AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft, OpenAI |
| Gold | Okta, Datadog, Docker, IBM, JetBrains, Salesforce, SAP, Shopify, Snowflake, Temporal, Twilio (among others) |
| Silver | Hugging Face, Pydantic, WorkOS, Uber, SUSE, Elastic, Cosmonic, Spectro Cloud (among others) |
How does AAIF stewardship differ from a normal Linux Foundation project?
AAIF is a directed fund, not a single project — a self-contained foundation-within-the-Linux-Foundation with its own governing board, technical committee, working groups and member tiers, hosting multiple projects rather than one. That is closer to how the CNCF hosts Kubernetes and dozens of cloud-native projects than to a standalone repo donated under the umbrella LF.
The practical difference for builders is durability and breadth. A directed fund has its own budget, its own dedicated staff, and a lifecycle policy for admitting and graduating projects — the Growth, Impact and Emeritus stages approved in 2026. That means AAIF is built to host the next wave of agent standards, not just to park the three founding donations. The foundation’s momentum has been striking: AAIF has been characterized as one of the fastest-growing efforts in Linux Foundation history, and its MCP Dev Summit North America in NYC drew roughly 1,200 attendees across 17 keynotes and 95-plus sessions in April 2026 — a doubling from the prior summit.
For decision-makers, the directed-fund structure is the part that de-risks adoption. You are not adopting a protocol that one company might abandon; you are adopting a protocol with a funded institution, a governance charter, a lifecycle policy, and a roomful of competitors who all agreed to share it.
On the technical roadmap, MCP’s published 2026 priorities — enterprise readiness (SSO-integrated auth, standardized audit trails), transport evolution, agent communication, and governance maturation — are now executed inside this structure rather than by one vendor. For the security implications of that enterprise push, see alatirok’s MCP security coverage.
Pros
Cons
What’s next for the Agentic AI Foundation in 2026?
AAIF is the governance upgrade MCP needed — and your code is safe
Expect AAIF to expand from a three-project donation into the default convergence point for open agent standards, with the agentgateway project, the new project lifecycle policy, and MCP’s enterprise-readiness roadmap as the near-term drivers. The summit signaled where the energy is going: gateways, gRPC transports, and observability that compose MCP, A2A and OpenTelemetry into something enterprises can govern at scale, per InfoQ’s coverage.
For builders, the strategic read is that the agentic AI foundation has shifted MCP from “promising standard” to “enterprise infrastructure.” The maintainers’ decision to keep MCP’s core narrow — and to push identity and policy into the gateway layer — tells you exactly where to invest your own engineering: the protocol is handled, the enforcement layer around it is the green field.
If you are building agent products, the action items are simple. Keep your MCP code as-is. Track SEPs through the unchanged maintainer process. Treat the gateway/policy layer as your differentiation surface, since AAIF deliberately left it out of core. And watch the lifecycle policy — the next standard your stack depends on may well be donated to this same fund.
Builder’s take
I build on MCP every day inside Cyntr and Loomfeed, so when Anthropic gave it away I read the charter, not the press release. Here is what actually matters if you ship agent code:
- Your MCP servers and clients do not change. Stewardship moved to a foundation; the spec did not fork and the wire protocol is identical. If you panicked and pinned versions, unpin.
- The governing board controls money and membership — not the spec. AWS chairs the board but the same maintainers still merge SEPs. Vendor logos buy a seat at the budget table, not commit rights.
- Neutrality is the whole point and it is a real moat. A single-vendor protocol is a procurement risk; a Linux Foundation directed fund is something an enterprise legal team will actually sign off on. That is why I stopped hedging MCP in Cyntr’s architecture docs.
- Watch the scope line, not the logos. Maintainers explicitly kept identity, observability and governance OUT of MCP core. That tells you where the gateway and policy layer (your job) begins.
- The three-stage lifecycle policy means more projects are coming. AAIF is now a destination, not a one-time donation. Expect A2A-adjacent and gateway projects to land here next.
Frequently asked questions
AAIF is a neutral, vendor-independent directed fund under the Linux Foundation that stewards open standards for AI agents. It launched December 9, 2025, co-founded by Anthropic, Block and OpenAI, and hosts the Model Context Protocol (MCP), goose, AGENTS.md and agentgateway.
No. The donation moved stewardship of MCP from Anthropic to AAIF, but the specification did not fork and the wire protocol is unchanged. Existing MCP servers and clients work exactly as before — there is nothing to migrate and no version to chase.
No single company owns MCP anymore. Stewardship sits with the Agentic AI Foundation under the Linux Foundation, while the same maintainers — led by core maintainer David Soria Parra — continue to set the roadmap through Spec Enhancement Proposals and working groups.
No. AAIF separates the governing board (which controls budget and membership, and is chaired by AWS) from project maintainers (who control the spec). A Platinum membership buys board influence over the foundation’s direction and spending, not commit rights or a vote on MCP Spec Enhancement Proposals.
AAIF hosts four projects: the Model Context Protocol (MCP) for connecting agents to tools and data, goose for running local-first agents, AGENTS.md for per-repository coding-agent instructions, and agentgateway, a policy and routing layer for MCP and agent traffic.
The eight Founding Platinum members are AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI. The Gold tier includes Okta, Datadog, Docker, IBM, JetBrains, Salesforce, SAP, Shopify, Snowflake, Temporal and Twilio, with Silver members including Hugging Face, Pydantic, WorkOS, Uber and SUSE.
Primary sources
- Linux Foundation Announces the Formation of the Agentic AI Foundation — Linux Foundation
- Donating the Model Context Protocol and establishing the Agentic AI Foundation — Anthropic
- OpenAI co-founds the Agentic AI Foundation under the Linux Foundation — OpenAI
- Block, Anthropic, and OpenAI Launch the Agentic AI Foundation — Block
- OpenAI, Anthropic, and Block join new Linux Foundation effort to standardize the AI agent era — TechCrunch
- Anthropic Donates the MCP Protocol to the Agentic AI Foundation — The New Stack
- MCP Dev Summit 2026: AAIF Sets A Clear Direction With Disciplined Guardrails — Futurum Group
- AAIF’s MCP Dev Summit: Gateways, gRPC, and Observability Signal Protocol Hardening — InfoQ
- Home – Agentic AI Foundation — Agentic AI Foundation
Last updated: June 6, 2026. Related: Governance.
