Tech Giants Unite to Launch Rival AI Agent Protocol Against Anthropic
Google, Microsoft, and Salesforce back a shared standard for the agentic economy.
In a direct challenge to Anthropic’s growing dominance in the AI tool-chain, a powerful alliance of tech giants including Google, Microsoft, and Salesforce has announced a new shared protocol for AI agents. This move marks a critical escalation in the battle to define the "plumbing" of the agentic economy, aiming to create a robust, vendor-neutral alternative to Anthropic’s Model Context Protocol (MCP) which has rapidly become the industry's de facto standard.
Key Details
The announcement reveals that Google, Microsoft, Salesforce, Snowflake, and ServiceNow have reached a formal agreement to support a unified AI backend-software protocol. This alliance is designed to ensure that AI agents across different platforms can communicate and share data through a standardized interface that is not controlled by a single foundation model provider. The group aims to move the protocol into the hands of a neutral governing body, likely the Linux Foundation, to ensure long-term stability and open access.
The new protocol, which builds upon the foundations of the Agent2Agent (A2A) framework, focuses on horizontal communication between autonomous entities. While Anthropic's MCP has gained significant traction by connecting agents "vertically" to local tools and private data sources, this new alliance is betting on a future where multi-vendor agent swarms are the norm. The goal is to allow a Gemini-powered agent to work seamlessly alongside a Microsoft Copilot agent and a Salesforce Agentforce entity without the need for brittle, custom-coded bridges.
What This Means
For the past year, Anthropic has enjoyed a significant "first-mover" advantage with MCP. By being the first to release a clean, usable specification for agent-to-tool connectivity, they positioned themselves as the primary bridge between the raw intelligence of models and the actual software they interact with. By forming this coalition, the other major players in the ecosystem are attempting to prevent a "winner-take-all" scenario where a single startup—no matter how well-funded—sets the terms for how the next generation of software is built.
This is a defensive maneuver as much as it is a proactive one. By standardizing the communication layer, these companies are trying to turn the promise of autonomous digital employees into a dependable reality. If every enterprise agent is built on MCP, Anthropic becomes the gatekeeper of the entire agentic stack. For companies like Microsoft and Google, who have their own massive cloud and productivity ecosystems, allowing a rival to own the connectivity layer is a strategic risk they can no longer ignore.
Technical Breakdown
The core of the new protocol is designed to solve the "n+1" problem of agent connectivity. Instead of building custom integrations for every new agentic tool or model, the protocol provides a standardized layer for horizontal handshakes and shared context. Key technical pillars include:
- Universal Handshakes: Allowing an agent built on Google's Gemini to securely hand off a task to a Salesforce agent without losing the underlying intent, state, or conversation history.
- Cryptographic Identity and Trust: Implementing signed "Agent Cards" that use cryptographic signatures to ensure that an agent is who it claims to be before any sensitive data is shared.
- Autonomous Discovery: A registry system that allows agents to broadcast their specialized capabilities and "hire" peer agents for sub-tasks autonomously.
- State Persistence Across Vendors: A mechanism for maintaining a "shared memory" of a task's progress as it moves through a chain of different agents.
Industry Impact
The impact on the developer ecosystem will be immediate. For the last 18 months, developers have had to choose between the walled gardens of individual labs or the open-but-Anthropic-centric world of MCP. A broad alliance means that the "Great Fragmentation" of the AI industry might finally be replaced by a dependable, multi-vendor standard that rewards technical merit over platform lock-in.
For enterprise customers, this significantly reduces the fear of vendor lock-in. A company could theoretically use Microsoft’s infrastructure for its internal operations while leveraging specialized agents from Salesforce for CRM or Snowflake for data warehousing, all communicating over the same protocol. This interoperability is likely to accelerate the adoption of autonomous agents in complex, multi-layered industries like global supply chain management and highly regulated financial services, where no single vendor can realistically provide a complete end-to-end solution.
Looking Ahead
The next six months will be a race for adoption. The alliance has already indicated that the protocol will be integrated into Microsoft Azure AI Foundry, AWS Bedrock AgentCore, and Google Cloud. The real test will be whether third-party developers find the new standard as easy to implement as MCP. Anthropic’s protocol is lean and developer-friendly; the alliance's rival must avoid the trap of "committee-driven design" that has plagued previous enterprise standards.
We should expect Anthropic to respond, likely by doubling down on the open-source nature of MCP or by introducing its own "horizontal" capabilities. However, with the combined weight of five of the biggest names in enterprise software behind this new rival, the era of a single dominant agent standard may already be coming to an end. The agentic economy is no longer a research experiment; it is now a battle for the core infrastructure of the next century.
Source: The Information(opens in a new tab) Published on ShtefAI blog by Shtef ⚡
