MCP Goes Neutral: Why the Foundation Move Matters

The Model Context Protocol launched in November 2024 as Anthropic's answer to a problem that was slowing down the entire agentic AI industry: there was no standard way for an AI model to connect to external tools and systems. Every integration was custom. Every connector was built by hand. Every organization that wanted their AI assistant to read a Notion page, query a database, or call an internal API had to build and maintain a bespoke bridge between their systems and the model. That work was expensive, brittle, and non-transferable.

MCP proposed a simple mental model: a universal protocol that defines how AI models discover and call tools, read resources, and receive context from external systems. The protocol is deliberately low-level — it specifies the communication contract, not the business logic. That design choice meant that anyone could implement a server for any system, and any MCP-compatible client could use any of those servers without custom integration work.

The adoption velocity from November 2024 to December 2025 was significant. Ten thousand active public servers inside twelve months is a meaningful signal for a new infrastructure protocol, especially considering that MCP servers are not consumer applications — they require technical implementation and are built by people with a real integration use case. The 10k figure undercounts private and internal implementations significantly.

Vendor-controlled protocols carry a structural problem that no amount of openness or good intentions fully resolves: the vendor controls the roadmap. When the company that created the protocol also ships the primary client that implements it, the protocol's evolution will always be influenced — consciously or not — by what serves that company's competitive interests. Enterprise technology buyers know this. It is why they treat single-vendor 'open' specifications with caution, and rightly so.

Foundation governance changes the calculus. An independent foundation with multi-stakeholder representation gives the protocol a decision-making structure that no single vendor controls. That means companies building integrations on MCP can do so with more confidence that the protocol will evolve in ways that serve the ecosystem broadly rather than any single participant. It also lowers the perceived switching cost risk — if Anthropic made a strategic pivot tomorrow, MCP would continue to exist and be maintained under independent governance.

The parallel to successful internet infrastructure standards is apt. HTTP, TCP/IP, and DNS were not built by one company and handed to the world after they proved useful. They were developed through open processes with broad participation, and that governance model is a significant part of why they became universal infrastructure rather than proprietary platforms. MCP is not at that scale yet, but the governance structure it now has is the correct foundation for that trajectory.

OpenAI's adoption of MCP in its Responses API earlier in 2025 was the first major signal that the protocol had genuine multi-vendor traction. Google's support for MCP natively in the Gemini CLI added a second major signal. Both of those decisions make considerably more sense in the context of a protocol moving toward neutral governance — joining an Anthropic-controlled specification carries different risk than joining an independent standard.

The practical question for enterprises is not whether MCP is interesting. It is whether building integration infrastructure on MCP is a durable investment or a bet on one vendor's staying power. Foundation governance substantially improves the answer to that question.

We have been advising clients to build their agent integration layer against MCP since early 2025, and this governance transition validates that direction. The risk we were asking clients to accept — 'build on a protocol that Anthropic controls' — has now changed to 'build on a protocol governed by an independent foundation with multi-vendor participation.' That is a meaningfully different risk profile.

The teams that invested in MCP-based connectors over the last year are already benefiting from reusability across tool surfaces that were not available when they started. A connector built for Claude works in any MCP-compatible client. As the foundation grows that roster of compatible clients, the leverage on that integration investment increases. The compounding here is real and it accelerates with foundation governance.

The immediate next step for the Agentic AI Foundation is likely to be specification refinement — addressing the rough edges in MCP's current version that practitioners have been working around. The protocol is functional today but it has known gaps in areas like authentication, authorization, and streaming that affect enterprise deployments specifically. Independent governance creates a better venue for those issues to be addressed systematically rather than through ad-hoc vendor updates.

Longer term, the question is whether MCP can achieve the kind of universal adoption that makes it true infrastructure — something you assume is present rather than something you choose to use. That requires adoption in more enterprise software platforms, more developer toolchains, and more operating environments. The foundation governance model is the precondition for that kind of broad adoption. The protocol being genuinely useful is the other precondition, and it has cleared that bar already.