Over the last year and a half, the reasoning models of the world’s leading AI labs have grown more sophisticated, better at adhering to instructions, and capable of completing multi-step processes. That has resulted in a surge in demand for AI agents, bots capable of autonomously executing 10-plus-step processes on behalf of users.
Agents are taking over the internet. Last year, automated traffic spiked 23.5% from the previous year, eight times the rate of human traffic growth, according to research from Human Security. Retail sites have seen especially high surges, with traffic from AI sources up almost 400% year-over-year in the first quarter of 2026, per Adobe.
Historically, agents weren’t able to connect to external servers or complete commerce transactions—the pipes simply weren’t there. To enable AI agents to connect to external tools and data sources without existing APIs or custom integrations, Anthropic in late 2024 established Model Context Protocol. MCP, overseen by the Linux Foundation, which has since become the de-facto technical standard for agent-to-tool compatibility; it’s the universal adapter that allows agents to move across the web and work across different environments.
In the year and a half since the debut of MCP, there’s been a proliferation of technical standards—many of which are enabled by MCP—designed to facilitate agentic advertising and commerce. Brands, agencies, platforms, and publishers have expressed growing interest in agentic advertising and commerce, which could theoretically enable simpler navigation of the fragmented programmatic ecosystem and lead to more efficient, transparent, and cost-effective transactions.
But the explosion of specs is a lot to keep up with. Here’s a simple guide to some of the leading technical standards and how they differ. Note that this list is by no means exhaustive and there is a longtail of ad and commerce standards emerging from various industry players.
The infrastructure layer
Model Context Protocol (MCP)
Rolled out in November 2024 by Anthropic and now governed by the Linux Foundation, MCP is the foundational standard that enables AI agents of all kinds to connect to external tech, data, and services. It works by linking an MCP client, which is typically an agent built on a major large language model, with local data sources or remote servers, allowing them to speak the same language.
Agent2Agent Protocol (A2A)
Introduced by Google in April of 2025, A2A is like MCP in that it is universally accessible and now managed by the Linux Foundation. However, rather than helping agents connect to external tech and data, A2A is a universal standard to agent-to-agent interactions. It allows agents to connect and communicate without exposing their internal logic, memory, or technical specs. A2A has gained widespread support from major SaaS players, including Microsoft, Amazon Web Services, SAP, Salesforce, and IBM. A2A was built atop a variety of popular standards including HTTP, SSE, and JSON-RPC.
