Agentic AI Will Make MCP the New HTTP — Organizations Without API-Exposed Content Will Be Invisible to the Next Generation of Buyers
The Claim
Model Context Protocol (MCP) will become as foundational to digital presence by 2028 as HTTP was in 2000. Organizations that do not expose their content, product data, and transactional capabilities through MCP and equivalent agent-to-agent protocols will be invisible to the AI agents that future customers will deploy to discover, evaluate, and transact on their behalf.
The Evidence for the Protocol Shift
MCP appeared in five separate sessions at EvolveDigital, an unusual level of consensus across independent presentations. Martin Anderson-Clutz framed it as an architecture requirement: websites must 'treat APIs as the new UI,' implementing well-documented agentic skills and supporting MCP so that downstream AI systems can participate in orchestration stacks. Andrew Kumar described MCP servers as enabling AI agents to work across different systems. Kevin Basarab demonstrated a live MCP implementation showing Claude interacting with content approval and metadata management workflows through a single interface. Preston So announced a React Bricks MCP server in active development.
Justin Cook's introduction of UCP — a consortium protocol backed by Google, Shopify, Target, and Walmart for enabling agentic commerce — suggests that the protocol-standardization momentum extends beyond MCP to the broader agent-to-agent communication layer. This is not fringe speculation; Walmart and Google are not early adopters in the experimental sense.
The 2028 Timeline Problem
The Gartner forecast that 80% of customer interactions will shift to agentic experiences by 2028 was cited by Andrew Kumar, but Gartner forecasts have a variable track record for specific timeline predictions. More significantly, no speaker at EvolveDigital presented measured business outcomes from MCP implementation at organizational scale. The demonstrations were working — Basarab's live demo was credible — but demonstrations and production deployment at organizational scale are different orders of magnitude.
Sean Stanleigh's explicit caution about AI hype cycles is directly relevant here. The 'new HTTP' analogy is rhetorically powerful but technically imprecise — HTTP succeeded in part because of its simplicity and its backing by Tim Berners-Lee, CERN, and later the IETF. MCP is a protocol from Anthropic. UCP is a consortium proposal. Protocol wars have long histories and uncertain outcomes.
What Organizations Should Actually Do
The directional argument is sound: begin exposing content through structured APIs, implement MCP where it fits existing workflows (Basarab's demo showed real value), and audit which content and transactional capabilities would be valuable to AI agents. These are prudent investments that do not require belief in the 2028 timeline.
The 'new HTTP' framing, however, deserves skepticism. Organizations that over-index on MCP-first architecture in 2026 before the protocol ecosystem stabilizes risk building on a fragmented foundation. Monitor, pilot, and implement where value is demonstrable — but do not bet the content architecture on a timeline that no one in the corpus has empirical evidence for.