Just over a year ago, MCP (Model Context Protocol) was a protocol released by Anthropic, circulating within a small developer community. Most people saw it as a technical specification for connecting AI tools, allowing Claude to interface with various external tools.
Now, it has appeared in a joint announcement from Google Cloud and ServiceNow. The use case: 5G network self-healing.
What was announced
On April 22, Google Cloud and ServiceNow announced a partnership at the Google Cloud Next conference, centered on the combined use of three protocols:
- MCP (Model Context Protocol): A standard interface for real-time data exchange between AI and external tools and data sources
- A2A (Agent-to-Agent): A protocol for AI agents to pass instructions and intelligence between each other
- A2UI (Agent-to-UI): A protocol for AI agents to directly manipulate user interfaces
What happens when these three protocols are stacked together?
Take 5G network operations as an example: Gemini's AI agent detects a network anomaly, uses MCP to retrieve real-time network data, coordinates multiple sub-agents via A2A to analyze the root cause, and finally triggers ServiceNow's automated workflow to complete the repair — all without human intervention.
Scattered tool calls have become true autonomous operations.
Where enterprise AI is really stuck
This matters because it touches a very practical pain point in enterprise AI deployment.
Is AI not smart enough? That problem was largely solved by 2024. What really holds enterprise AI back is cross-system collaboration.
A typical large company's IT environment looks like this: network operations run on ServiceNow, data lives in BigQuery, models run on Gemini, and security policies are managed by yet another system. Each system is an island, and AI capabilities are locked inside, unable to escape.
What MCP does is build a unified road between these islands.
A2A solves a harder problem: the "language" issue when multiple AI agents collaborate. Different agents running on different systems need a common communication format to truly work together, rather than operating in silos.
Zero-copy data: Another detail
The announcement also mentioned "Workflow Data Fabric," using BigQuery ML's zero-copy data connection.
This detail is important. A common pitfall in enterprise AI: to let AI analyze data, you usually need to copy the data into the AI system first. This is slow and carries security and compliance risks — who is responsible for that copied data?
Zero-copy means AI accesses data directly where it resides, rather than moving it. For data-sensitive industries like finance and healthcare, this architecture matters more than model capability itself.
Another signal: Partner awards
Google Cloud awarded ServiceNow the 2026 Partner of the Year award in four categories simultaneously, including one specifically called "Agentic AI Innovation."
This is the first time Google has created a dedicated partner award category for agentic AI.
The creation of this award category itself is a signal: Google believes agentic AI has reached a stage where its business value can be independently evaluated, no longer just an accessory to model capabilities.
The path MCP has traveled
From Anthropic's release of MCP to now, the trajectory looks roughly like this:
- AI tool ecosystem standard: Enabling tools like Cursor and Claude Code to connect to external data and systems
- Linux Foundation takes over: Becoming industry infrastructure, no longer just Anthropic's proprietary protocol
- Google Cloud enterprise deployment: Applied in production scenarios like 5G operations and retail predictive maintenance
This path is somewhat reminiscent of early HTTP — expanding from research networks to the commercial internet, becoming a foundational layer that everyone uses but no one particularly notices.
The scale and importance are not comparable, but the logic of diffusion is similar.
MCP now carries two types of demands simultaneously: AI tool integration on the developer side, and autonomous operations on the enterprise side. The users and requirements of these two scenarios are completely different.
Whether it can excel at both is the real test for MCP to become a true standard.
Sources: CocoLoop, ServiceNow and Google Cloud Unite AI Agents for Autonomous Enterprise Operations (Google Cloud Press Corner)