Agent Standards War: What the MCP Battle Means for Swiss SMEs

In mid-July 2026, Google, Microsoft, Salesforce, Snowflake and ServiceNow formed an alliance to develop their own standard for connecting AI agents to business software. The move is a direct response to Anthropic's Model Context Protocol (MCP), which has effectively become the industry standard with roughly 97 million monthly SDK downloads. For decision-makers at Swiss companies, this is not a footnote from the tech press — it is the question of what foundation your entire agent infrastructure will be built on for years to come.

What happened: Five major vendors against an open standard

On July 17, 2026, The Information first reported on the new alliance, which Info-Tech Research Group confirmed and contextualised in its 'Big 5 AI Vendor Roundup' report published July 23. The alliance is notable because it brings together five companies that otherwise compete directly across almost every market segment — cloud, CRM, data platforms, workflow automation. Their common denominator: together they hold the majority of the world's business data, and they want to prevent the agent era from being built on infrastructure controlled by an outside vendor.

MCP: How an open standard became critical infrastructure

Anthropic released the Model Context Protocol in December 2024 as an open-source project — a unified protocol through which AI agents can communicate with external tools, databases and applications, independent of the underlying model. In December 2025, Anthropic transferred governance to the Linux Foundation, specifically the newly established Agentic AI Foundation — a clear signal that MCP was meant to become neutral, open infrastructure rather than remain a proprietary Anthropic product. The strategy has worked: every major AI lab has since adopted MCP.

Why competitors are joining forces

The logic behind the Big 5 alliance is straightforward: as agents increasingly execute tasks autonomously across CRM, ERP, cloud storage and workflow tools, the underlying protocol determines who retains control over the data flow. A standard controlled by an outside lab — even if formally housed under a neutral foundation — shifts the balance of power away from the platforms that traditionally own the customer relationship. Google, Microsoft, Salesforce, Snowflake and ServiceNow are not willing to accept that shift without a fight.

A familiar dynamic: browsers, USB, cloud

The situation echoes earlier standards wars. In the browser wars of the 1990s, Netscape and Microsoft fought not over the better product but over whose technology would become the web standard. With USB-C versus proprietary connectors, the battle was about compatibility and control over the ecosystem of connected devices. With AWS, Azure and Google Cloud, the chosen platform still determines migration costs and dependencies today. The agent standards question follows the same pattern — except this time, five vendors are jointly pushing back against an open standard.

• Browser wars: control over the web standard, not the better product
• USB vs. proprietary connectors: compatibility as a power question
• Cloud standards (AWS/Azure/GCP): migration costs as a lock-in mechanism
• Agent standards 2026: control over data flow between agents and business software

What this means strategically for Swiss SMEs

For Swiss companies deploying or planning to deploy AI agents in production, the central question shifts from 'which model do we use' to 'which ecosystem do we build on'. As agents increasingly replace classic enterprise software, the choice of underlying infrastructure becomes a strategic decision rather than an IT footnote.

For Swiss companies, an additional dimension applies: the revised Data Protection Act and the question of data sovereignty. Committing to a Big 5 ecosystem today implicitly determines where and how agent data is processed, logged and shared. An open, neutral standard such as MCP tends to give you more control over which components of your agent infrastructure you operate or audit yourself — a proprietary Big 5 solution binds you more tightly to that vendor's governance rules.

Multi-model becomes multi-standard strategy

Until now, the pragmatic recommendation for companies was to avoid relying on a single language model and instead stay flexible. The Agent Standards War adds a second layer: the choice of orchestration and connection layer should also not be locked into a single, closed ecosystem. Any decision about which platform to use for production agent deployment should explicitly examine whether that platform remains MCP-compatible or ties you exclusively to one of the Big 5 standards.

This question is not purely academic. Many projects already fail at the jump from pilot to production. A wrong standard choice at the pilot stage compounds this risk further, because later migrations consume time and budget that should otherwise go toward scaling.

Concrete recommendations for Swiss decision-makers

1. Explicitly verify for every new agent investment whether the solution supports MCP or locks you exclusively into a Big 5 ecosystem.
2. Avoid contracts and architectures that make migration between standards technically or contractually difficult.
3. Treat the standards question as a board-level matter, not an IT detail — it affects cost, compliance and strategic flexibility.
4. Actively monitor how the Big 5 alliance develops; consolidation around a shared, interoperable standard remains possible.
5. Build in multi-standard capability rather than committing early to a single ecosystem.

Conclusion: The standards question belongs in the boardroom

The Agent Standards War shows that competition in the AI industry has shifted from models to the infrastructure on which agents operate. For Swiss companies, this means the decision of which ecosystem to trust deserves the same strategic diligence as choosing an ERP system decades ago — with the difference that the market now moves considerably faster, and open, neutral standards like MCP preserve more negotiating leverage than an early commitment to a closed ecosystem.