On December 9, 2025, the Linux Foundation announced the Agentic AI Foundation (AAIF), a new home for open standards and tools for AI agents. Anthropic donated the Model Context Protocol (MCP), Block contributed its open-source agent goose, and OpenAI contributed AGENTS.md, a convention for giving coding agents project instructions. Moving these projects under a neutral foundation matters because it lowers the risk of building on them: no single AI vendor now controls the protocol that most agent tooling depends on.

Key Facts#

  • Date: December 9, 2025, announced by the Linux Foundation, the MCP project and Anthropic.
  • Structure: The AAIF is a directed fund under the Linux Foundation, with a governing board that handles strategy, budget, membership and project approvals.
  • Founding projects: MCP (donated by Anthropic), goose (contributed by Block) and AGENTS.md (contributed by OpenAI).
  • Co-founders: Anthropic, Block and OpenAI, with Google, Microsoft, Amazon Web Services, Cloudflare and Bloomberg named as supporting members.
  • MCP adoption at donation: More than 10,000 active public MCP servers and more than 97 million monthly SDK downloads across Python and TypeScript, according to Anthropic.
  • Governance: MCP's maintainers keep their existing structure and decision process, including the SEP process used to propose specification changes.

What Happened#

The announcement combined three projects that address different layers of agent development. MCP defines how an AI application connects to external tools and data. goose is a working agent that uses MCP to reach those tools. AGENTS.md tells coding agents how to work inside a particular repository. Placing them in one foundation gave the ecosystem a shared, vendor-neutral venue for the standards that sit between models and the software they operate.

MCP's lead maintainer, David Soria Parra, wrote on the MCP blog that the move gives the protocol "the same neutral stewardship that supports Kubernetes, PyTorch, and Node.js." He also stressed continuity. The foundation's governing board decides on funding, membership and which projects join, while MCP's maintainers retain full autonomy over technical direction and day-to-day operations. Anthropic made the same point in its own post, saying the project's governance model would remain unchanged.

Adoption figures explained why neutrality had become urgent. Anthropic reported more than 10,000 active public MCP servers and noted that ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code had all adopted the protocol. Its Claude directory alone listed more than 75 connectors built on MCP.

The two other founding projects are smaller but useful. goose, released under the Apache 2.0 license, is an open-source agent that runs locally as a desktop app for macOS, Linux and Windows, as a command-line tool, or through an API. Its README says it works with more than 15 model providers and connects to more than 70 extensions through MCP. AGENTS.md is a plain Markdown file that its maintainers describe as "a README for agents": a predictable place to put build commands, testing steps and contribution rules for coding agents.

Background#

Anthropic introduced MCP in November 2024 as an open specification with SDKs and reference servers. Through 2025 the protocol went from a single vendor's proposal to an industry default. OpenAI adopted it in March, Google shipped MCP support in Gemini CLI in June, and Microsoft built it into its developer tools. The project also matured institutionally, publishing a formal governance model in July and launching the MCP Registry in preview in September.

That growth created a governance tension. Competitors were building products on a protocol that one of them had created and still stewarded. Anthropic had maintained that MCP would stay open and community-driven, but a foundation turns that promise into a structure. The Linux Foundation model, which already hosts projects such as Kubernetes and Node.js, separates funding and membership decisions from technical governance.

For background on the adoption wave, see our story on OpenAI adopting MCP.

Why It Matters for Developers#

For .NET teams, the main effect is lower platform risk. The official C# SDK lives in the MCP project's GitHub organization, so a server or client built today depends on a protocol whose governance no longer rests with one company. That makes MCP a safer long-term choice for internal tool integration. Our MCP in C# guide and the story on the C# SDK 1.0 release cover the practical side.

AGENTS.md is the most immediately useful project for many teams, because it costs one file. Coding agents make fewer mistakes when a repository states its commands and rules explicitly. A .NET repository might start with something like this:

Text
# AGENTS.md

## Build and test
- Restore and build: dotnet build -c Release
- Run all tests: dotnet test --no-build -c Release
- Format before committing: dotnet format

## Conventions
- Target net10.0; nullable reference types are enabled and warnings are errors.
- Put new endpoints in src/Api/Endpoints and add tests in tests/Api.Tests.
- Never edit files under src/Generated; they are produced by source generators.

## Pull requests
- Keep changes small and include tests for new behavior.

A file like this works best when the build itself enforces the rules it describes. Analyzers, Directory.Build.props settings and warnings-as-errors turn prose instructions into checks an agent cannot ignore, as our code quality guide explains. For a broader view of working with agents on .NET codebases, see our guide to AI-assisted .NET development.

goose is also worth a look for teams that want an open, local agent they can inspect. Because it speaks MCP, it can use the same .NET MCP servers you build for other clients.

What's Next#

The foundation's first job was to keep the founding projects moving without disruption, and MCP continued to publish specifications and extensions under its existing process. The AAIF then built out an events program. It scheduled MCP Dev Summit North America for April 2 and 3, 2026 in New York City, with a program of more than 95 sessions, and announced AGNTCon + MCPCon Europe for September 17 and 18, 2026 in Amsterdam, an AGNTCon + MCPCon North America event in October, and regional MCP Dev Summits in cities including Mumbai, Seoul, Shanghai, Tokyo and Toronto.

Open questions include how the foundation will decide which additional projects to accept, how it balances large corporate members against independent maintainers, and whether other agent standards will consolidate under its umbrella or remain separate.

Sources#