The Protocol Landscape
A protocol is an agreed format and process for exchange between systems. Two computers that need to communicate do not negotiate their communication rules from scratch; they implement a shared protocol, and the shared protocol makes interoperability possible without custom integration for every pair of systems.
In AI-native advertising, protocols perform the same function at the level of agents, supply descriptors, and intelligence systems. They define how buyer and seller agents discover each other, exchange supply and demand signals, negotiate terms, and report outcomes — in standard formats that allow different vendors' systems to work together without proprietary lock-in.
There are three protocols worth understanding for anyone building or operating in this space: AdCP as the advertising-specific protocol layer, and MCP and A2A as general-purpose agentic infrastructure that AdCP and similar standards build upon.
AdCP — The Ad Context Protocol
AdCP (the Ad Context Protocol) is an active, open, versioned protocol for agentic advertising. It is evolving — not a finished universal standard — but it defines a shared vocabulary and interaction model that buyer agents and seller agents can implement today.
AdCP operates at the advertising-specific protocol layer. It governs how buyer and seller agents interact across the full campaign lifecycle: discovery (finding eligible supply), planning (evaluating fit against campaign intent), buying and selling (negotiating and executing placements), and measurement (reporting outcomes). Supply descriptors and provenance records are part of what AdCP carries, but AdCP is not limited to supply description — it is the protocol through which the agent handoff happens.
AdCP is designed to be:
Declarative. Agents and descriptors express supply properties, campaign requirements, and negotiation terms as structured fields rather than free text that each party must parse independently.
Extensible. The protocol accommodates additional fields and metadata across versions without breaking existing implementations. As the ecosystem evolves, AdCP can carry richer information while remaining backward-compatible.
Provenance-aware. AdCP includes formats for attaching chain-of-custody information to supply claims, enabling buyer agents to evaluate the reliability of assertions alongside the assertions themselves.
AdCP is one protocol in a forming landscape; it is not the only possible standard for agentic advertising. The important concept is that open, versioned protocols exist, can be inspected, and enable interoperability — not that any single protocol is the definitive answer.
MCP — The Model Context Protocol
MCP (the Model Context Protocol) is an open protocol developed by Anthropic that standardizes how AI models access external tools and data sources. It operates at a different layer than AdCP — not at the advertising-specific layer, but at the interface between an AI model and the tools it uses.
An AI buyer agent is a model — a large language model or a purpose-built AI system — that needs access to tools to do its job: querying supply descriptors, reading campaign briefs, evaluating inventory, executing buys. MCP defines a standard way for models to discover and use these tools, regardless of who built the model or who built the tools.
Without MCP, every AI model and every tool need custom integration. With MCP, a tool that exposes an MCP-compliant interface is accessible to any model that speaks MCP — reducing integration complexity across the agentic stack.
In an AI-native advertising system, a buyer agent might use MCP to access tools that query supply intelligence systems or initiate AdCP-formatted negotiations. MCP is the general infrastructure layer; AdCP is the advertising-specific protocol those tools may implement.
In practice
A media agency builds a buyer agent using a commercial LLM. They want the agent to discover supply and negotiate placements through AdCP. Rather than building a custom integration for every supply intelligence provider, the agency's agent uses MCP to discover and call tools that speak AdCP. The agent speaks MCP to reach the tools; the tools speak AdCP to conduct the advertising interaction. Neither party needed to build a proprietary connector for every pairing.
A2A — The Agent-to-Agent Protocol
A2A (Agent-to-Agent) is an open protocol developed by Google that standardizes communication between AI agents. Where MCP addresses the interface between a single agent and its tools, A2A addresses the interface between multiple agents operating in coordination.
In AI-native advertising, the relevant agent interactions are between buyer agents and seller agents — the negotiation exchange described in Unit 3. A2A provides a general vocabulary for those interactions: how agents announce their capabilities, how they exchange structured requests and responses, and how they maintain context across a multi-turn exchange.
AdCP applies A2A-style principles to the specific domain of advertising — discovery, planning, buying, selling, and measurement between buyer and seller agents. A2A is the general infrastructure; AdCP is the advertising-specific protocol built for that infrastructure.
A Multi-Protocol Stack
The three protocols described here operate at different layers and address different problems:
| Protocol | Layer | Function | |---|---|---| | AdCP | Advertising-specific | Buyer/seller agent interaction across discovery, planning, buying, selling, and measurement | | MCP | Model tooling | Standard interface between AI models and the tools they use | | A2A | Agent communication | General protocol for agent-to-agent interaction and coordination |
These protocols are complementary rather than competing. A complete AI-native advertising system might use all three: MCP to give buyer agents standard access to tools, A2A principles for general agent coordination, and AdCP for the advertising-specific negotiation and transaction layer.
The protocol landscape is still early. Standards are evolving, implementations vary, and not all parties in the ecosystem have adopted the same set of protocols. The important principle is protocol agnosticism: building systems that are not locked to a single protocol or vendor, and that can operate across protocol versions as the ecosystem matures.
From Foundations to Practice
This unit is the last in the Foundations sequence. You have covered, in order: why media buying is shifting to machine systems, what AI-native advertising means architecturally, how buyer agents and seller agents work, what machine-readable supply requires, what supply descriptors contain, how supply intelligence evaluates them, and how provenance makes the evaluation trustworthy. This unit completes the picture with the protocol layer that connects the pieces.
The Foundations course was designed as a mental model, not a technical specification. Deeper technical treatment of AdCP, supply intelligence architecture, and the agent interaction model will appear in future Learn sections as they are published. The Glossary contains definitions for every term introduced in these eight units.
The goal of these foundations is not memorization. It is orientation: a clear enough picture of how the system works that you can evaluate what you encounter in this space with confidence about where things fit and what they actually do.