MCP, A2A, ARD, UCP, WebMCP, OKF: Six Protocols Solving Six Different Problems in the Agent Stack

The agent infrastructure ecosystem now has at least six named protocols, and most conversations about them get the framing wrong. According to a Search Engine Journal analysis by Chris Green, Technical Director at Torque Partnership, these standards are not competing alternatives. They operate at different layers of the agent stack, solving different problems for different use cases.

The Six Layers

Green maps the protocols across two axes: action-oriented versus knowledge-oriented, and agent-driven versus publisher-driven. The result is a taxonomy where overlap exists but outright competition is rare.

MCP (Model Context Protocol) handles tool invocation. When an agent needs to call an API, query a database, or interact with a SaaS application, MCP defines how the agent discovers and uses those capabilities. It operates at the agent-tool binding layer. Anthropic introduced MCP in late 2024, and it has become the most widely adopted protocol in the stack, supported across Claude, OpenClaw, OpenAI’s Codex, and dozens of agent frameworks.

A2A (Agent-to-Agent) handles coordination between agents. When one agent needs to delegate work to another, A2A defines the handoff protocol. Google launched A2A as an open standard in early 2025. A2A’s creators have positioned it as sometimes preferable to MCP server configurations in multi-agent orchestration scenarios, but Green’s analysis treats them as solving adjacent problems rather than competing directly.

ARD (Agent Resource Discovery) handles capability advertisement. When an agent needs to find what actions a website or service supports, ARD provides the discovery mechanism. It sits upstream of MCP: ARD tells agents what exists, MCP tells them how to use it.

WebMCP extends the MCP concept to web-based agent actions. Where MCP typically operates through API connections, WebMCP targets scenarios where agents interact with websites directly, exposing page-level capabilities to agent frameworks.

UCP (Universal Commerce Protocol) targets ecommerce specifically. Product discovery, cart construction, checkout flows: UCP defines how agents complete purchase transactions. For retailers, this is likely the first agent protocol they will encounter in production, according to Green’s analysis.

OKF (Open Knowledge Framework) sits at the knowledge layer. For large, complex websites with deep content hierarchies, OKF helps AI systems discover and understand content structure at scale.

Where the Overlap Actually Is

Green identifies the overlap zone as discovery, invocation, and orchestration. ARD helps agents find capabilities. MCP and WebMCP help expose or invoke tools. A2A helps agents coordinate. UCP applies these patterns to commerce journeys. The protocols layer rather than compete.

This framing matters because the industry conversation has defaulted to a “which protocol wins” narrative. Green’s analysis suggests the answer is closer to “all of them, at different layers,” similar to how web browsers rely on multiple HTML-related standards simultaneously rather than choosing a single specification.

The Priority Problem

Green identifies a common pattern: most organizations are worrying about step four (monitoring emerging standards) while struggling with steps one and two (understanding the landscape and improving discoverability). The pragmatic recommendation: if you’re in ecommerce, watch UCP. If you expect agents to complete user-facing actions, monitor WebMCP and ARD. If you manage complex content architectures, track OKF.

For agent builders specifically, MCP remains the protocol with the most immediate impact. It has crossed from technical specification to enterprise procurement requirement, with A2A following close behind. The others are earlier in adoption but address real gaps that MCP alone does not fill.

One Ecosystem, Not One Standard

The uncomfortable reality for teams evaluating agent infrastructure: there will not be a single winner. Standards become standards through ecosystem adoption, not technical superiority, and the agent stack has enough distinct layers to sustain multiple protocols simultaneously. The real risk is waiting for protocol convergence that may never arrive while competitors ship agents that already interoperate across multiple standards.