MCP Weekly: The Agentic Explosion Hits Commerce, Work, and Cloud

Annnd we’re back! Welcome to the first MCP Weekly for 2026, covering major developments from January 1 through January 14. The trend over these last two weeks was obvious: commerce, work, and cloud are being rebuilt around agents that actually do things. UCP, Copilot Checkout, and the new Slackbot are all bets on this same idea: that the next competitive surface is going to be the protocol layer. The hard problem is proving intent, enforcing limits, and keeping actions safe.

TL;DR

Google and Shopify launched the Universal Commerce Protocol (UCP), scaling retail AI tokens 11x to 90 trillion. Microsoft countered with Copilot Checkout, grounding agents in real-time inventory via MCP. Meanwhile, Anthropic’s Cowork and Salesforce’s Slackbot shifted the focus to autonomous, local file agency, and enterprise-wide "Work OS" automation.

Google’s Universal Commerce Protocol and the 90-Trillion-Token Moment

The headline of the week belongs to the National Retail Federation (NRF) conference, where Google unveiled the Universal Commerce Protocol (UCP). The data is staggering: in December 2024, Google processed 8.3 trillion retail tokens; by December 2025, that number hit 90 trillion, an 11X explosion.

• What is UCP?: An open-source framework co-developed with Shopify and Walmart. It provides a "common language" for AI agents to handle the entire journey from discovery to post-purchase support.
• Interoperability: UCP is built on top of MCP (Model Context Protocol), AP2 (Agent Payments Protocol), and A2A (Agent2Agent).

Microsoft Grounds Commerce with MCP-Powered Checkout

Microsoft responded by launching Copilot Checkout, a native commerce interface that allows shoppers to browse and buy without leaving the chat.

• MCP Integration: Copilot Studio now supports direct connection to MCP servers, allowing agents to ground their reasoning in real-time inventory and pricing data.
• From Static Listings to Agentic Storefronts: Shopify merchants are being automatically enrolled, transforming static listings into interactive, transactional agents.

The Emergence of the Agentic Work OS

• Anthropic "Cowork": A research preview that adapts the Claude Agent SDK for non-technical desktop workflows. Claude is granted autonomous read/write access to local folders, isolated in an Apple Virtualization sandbox, allowing it to synthesize reports and manage files with high agency.
• Slackbot Relaunched: Salesforce has turned Slack into an "Agentic Work OS". The new Slackbot (powered by Agentforce 360) synthesizes context from your DMs, channels, and connected systems like Google Drive and M365 to automate "last mile" tasks without explicit prompting.
• Anthropic “Labs”: Anthropic has announced the expansion of Labs, a dedicated team focused on incubating experimental products at the frontier of Claude’s agentic capabilities.

My Thoughts: The Rise of the Protocol Architect

Model quality is converging fast. That means “intelligence” is becoming table stakes. The real differentiator is shifting to something more operational: agency you can trust.

This week is a good snapshot of that shift. UCP is pushing commerce toward a shared transaction surface. Copilot Checkout brings purchasing into the chat layer, but only works if agents can ground themselves in live inventory and pricing through MCP. Slackbot and Anthropic Cowork take the same idea into work, where the value is not a better answer, it’s reliable action across files, messages, and systems.

The common thread is that we are leaving the experimentation phase. When Slackbot claims 20 hours a week saved, or when agents drive up to 60% higher return on ad spend, the bottleneck stops being prompts. It becomes permissions, auditability, failure modes, and cryptographic proof that a human actually approved what happened.

That’s why I think the “protocol architect” role is about to matter a lot. In the agent era, distribution flows through whatever agents can transact with safely. If your brand is not UCP compatible, you are harder to reach, harder to compare, and eventually easier to ignore. The next wave of competitive advantage will come from building systems that are easy for agents to use, with clear limits and verifiable trust baked in.

‍