What the Linux Foundation and AAIF Mean for MCP and Agentic AI

Linux Foundation Agentic AI brings the central protocols of the agent world together under vendor-neutral, open governance. Anthropic handed the Model Context Protocol (MCP) to the newly founded Agentic AI Foundation (AAIF) under the umbrella of the Linux Foundation on 9 December 2025. Google had previously handed its Agent2Agent protocol (A2A) to the Linux Foundation back on 23 June 2025. As a result, the two de facto industry standards for agent interoperability now rest in independent hands rather than under the control of a single corporation.

• Who handed over what? Anthropic gave MCP to the AAIF (with Block and OpenAI as co-founders), and Google gave A2A directly to the Linux Foundation (founding members: AWS, Cisco, Google, Microsoft, Salesforce, SAP, ServiceNow).
• What does it bring enterprises? Vendor neutrality, open co-determination and therefore investment security: anyone relying on MCP + A2A builds on the convergent standard layer, not on a proprietary island.
• What remains open? Lock-in can still arise via the platform layer and via proprietary protocol extensions; this needs to be secured contractually.

Why the handover is the turning point for open agent standards

Until mid-2025, practically every multi-agent project wrote its own inter-agent communication. By 2026, two protocols have established themselves as the industry standard: MCP for connecting agents to tools, data and systems (agent-to-tool) and A2A for collaboration between independent agents across trust boundaries (agent-to-agent). The official reading shared by Google, Salesforce and Microsoft is: MCP for capabilities, A2A for collaboration — both are complementary, not competing.

The decisive step was that these protocols left their originating vendors. A protocol owned by a single vendor is always a latent lock-in risk: the owner can unilaterally change licensing terms, the roadmap or interfaces. The handover to the Linux Foundation neutralises this lever and turns a vendor asset into a shared, openly governed infrastructure.

The AAIF and A2A governance in detail

The Agentic AI Foundation (AAIF) is a directed fund under the Linux Foundation. It was co-founded by Anthropic, Block and OpenAI. MCP moved there alongside Block's goose and OpenAI's AGENTS.md. Important for the governance assessment: ongoing technical maintenance stays with the existing MCP maintainers; the AAIF governing board is responsible for strategic investments and the acquisition of further members. There is therefore a clean separation between technical development and strategic steering.

The A2A project was announced on 23 June 2025 at the Open Source Summit North America in Denver. Google donated the specification, the SDKs (Python, TypeScript) and the developer tooling. More than 100 companies support the protocol. IBM's competing ACP protocol was merged into A2A under LF AI & Data on 29 August 2025 — the ACP team ceased active development and transferred technology and expertise into A2A. The identity and observability project AGNTCY (Cisco/LangChain/Galileo) was also handed to the Linux Foundation on 29 July 2025. The result is not a vendor war but a coherent open protocol stack.

Who stands behind the standards (as of 2026)

The breadth of the sponsorship is the real signal for vendor neutrality. When direct competitors jointly back a protocol, none of them can appropriate it alone.

The fact that AWS, Google and Microsoft — as direct cloud competitors — are jointly Platinum supporters of MCP and at the same time A2A founding members is exactly the constellation that makes a standard viable and hard to capture.

What this means in concrete terms for technology choice

For DACH decision-makers, standardisation shifts the central question. In 2026 it is no longer "which proprietary agent ecosystem do I choose and tie myself to", but "how do I ensure that my platforms speak MCP and A2A cleanly".

• Investment security: An agent that speaks MCP and A2A remains interoperable even if you later switch the underlying platform or model. The standard layer outlasts individual product decisions.
• Reduced lock-in at the protocol layer: No vendor can unilaterally close the interface. This shifts competition from "who owns the interface" to "who delivers the best implementation".
• Realistic residual risk: Lock-in has not disappeared, it has shifted. It still arises at the platform layer (e.g. native orchestration in Salesforce Agentforce, SAP Joule Studio or Microsoft Agent 365) and through proprietary extensions. The most important watchlist question for 2026 is whether the AgentCard — the JSON document with which an A2A agent publishes its capabilities — stays a portable standard document or becomes fragmented by vendor-specific additions. Every founding vendor has commercial incentives to differentiate itself through such "A2A-plus" extensions.

Practical example: cross-vendor workflow with investment protection

A typical enterprise workflow in 2026 runs across several vendor domains: a CRM agent in Salesforce, an ERP agent in SAP, a productivity agent in Microsoft. The open standard layer makes this composition possible without chaining yourself to a single vendor:

```text
Salesforce agent --A2A--> SAP Joule agent --A2A--> Microsoft Agent 365
| | |
MCP tools MCP tools MCP tools
(CRM data) (ABAP, HANA) (calendar, files)
```

Microsoft's own multi-agent guidance puts it unambiguously: "Use MCP for tool and data access. Use Linux Foundation A2A for cross-platform agent-to-agent messaging." The A2A mechanism keeps each agent's internal prompts, models and memory opaque — one domain calls another without exposing its inner workings. It is precisely this property that makes cross-vendor workflows governable in the first place.

The consistent rollout sequence is: standardise the MCP server policy → first introduce A2A within one platform domain → extend to a second domain → publish an AgentCard directory → only add cross-vendor identity (e.g. via AGNTCY) once the volume justifies it. AGNTCY and the research-oriented NANDA belong on the 2026 watchlist, not on the critical path.

Contractual consequence for procurement

Open standards only protect when they are also demanded. The central procurement question for 2026 is: Does the Agentforce, Joule Studio or Copilot Studio contract guarantee A2A interoperability and MCP standards compliance — and what happens on exit? At a minimum, you should contractually secure: guaranteed A2A interoperability, an obligation to publish portable AgentCards, MCP conformity for tool connections, and clear exit and data portability clauses. All version, membership and date details here apply as of 2026 and subject to further developments.

For agencies and B2B decision-makers

For marketing agencies and AI-native service providers, the open standard layer is the biggest product lever: anyone who publishes an AgentCard for every productised agent offering makes their services callable from customer estates (Agentforce, Joule, Copilot Studio) without custom integration — and that scales across clients. For B2B decision-makers in the mid-market, the rational default for 2026 is clear: n8n or LangGraph for orchestration, MCP for tools, A2A for every cross-platform handshake. Blck Alpaca from Vienna supports DACH companies in aligning their agent strategy with these open standards, defusing lock-in risks contractually and building investment-secure multi-agent architectures. Get in touch if you want to secure your technology choice against the convergent MCP-plus-A2A layer.