OpenRTB. Prebid. Now AdCP: The Standard That Lets AI Run Your Media

New Industry Initiative: Ad Context Protocol (AdCP) Sets Open Standard for AI-Powered Advertising Automation

On October 15th at 11:00 AM EST, a consortium comprising publishers, platforms, agencies, and independent technology firms officially launched an autonomous foundation – Ad Context Protocol (AdCP). AdCP is an open standard for advertising automation that facilitates natural language interaction between AI assistants and advertising platforms, built upon the Model Context Protocol (MCP) for native agent interoperability. The RFC/v0.1 specification is currently available on GitHub. Notable participants include Optable, PubMatic, Scope3, Swivel, Triton, Adgent, Bidclick, Classify, Kiln, Locala, OpenAds, The Product Council, and Samba TV.

OpenRTB standardized how bids move. Prebid standardized how demand is stitched. The next wave standardizes how work gets done. The Ad Context Protocol (AdCP) is an open standard that lets AI agents discover data, plan, activate, and measure across fragmented ad stacks through one interface – think of it as the universal, OS agnostic USB-C of advertising workflows.

What AdCP is (definition, not hype)

AdCP defines agent-friendly interfaces (built on the Model Context Protocol, MCP) so assistants can operate common tasks e.g., discovering and activating audience/context signals, coordinating platform actions, and returning human-readable rationales. Its Signals Activation Protocol is a concrete example: an MCP schema for agents to find, rank, and deploy segments with provenance, freshness, performance, and cost metadata, all expressed in natural language and structured responses.

Why it matters now

APIs solved connectivity, not composability. Every brand still hand-codes brittle glue across SSPs, DSPs, clean rooms, ID graphs, and measurement. AdCP collapses those bespoke layers into a common agent interface so the same instruction – “Tell the agent to reach more new people without bombarding the same ones, while sticking to lower-emissions ad supply. It finds good inventory, suggests small budget shifts, drafts the changes, and logs what it did and why.” – can execute across platforms and report back with explainable steps and transparent rationale. The result: faster integrations, portable automations, and less vendor lock-in.

From OpenRTB & Prebid to AdCP: a useful analogy

• OpenRTB gave us a lingua franca for auctions.

• Prebid organized demand selection and fairness in the page/app runtime.

• AdCP organizes the operational loop-define → discover → compare → activate → measure – so agents can execute cross-platform tasks with consistent semantics. It’s the missing layer that turns discrete APIs into an automation fabric for planning, activation, and analytics.

Agentic AI as a “currency”

In my DMEXCO column on context-based agents, I argued that persistent “context agents” become living intelligence profiles that plan and act, not just classify. AdCP turns that idea into a transactable unit of work: an agent can request signals, justify trade-offs, and deploy changes with auditable steps. As MCP and Google’s Agent-to-Agent (A2A) gain traction, MCP↔MCP and A2A interop make these units portable across stacks – effectively a currency of actions rather than just data.

Concretely, an AdCP “get_signals” call returns ranked options with deployment status and recommendations; an “activate” task pushes the winners to DSPs/SSPs/CDPs, then logs outcomes back to the agent memory – closing the loop your team would otherwise close manually.

What becomes possible (examples)

• Clean-room to activation, agent-driven: An agent queries available segments (coverage, recency, price), selects best performance audiences, activates to DV360, and posts the rationale in Slack – same flow across partners.

• Cross-platform governance by design: Provenance and freshness live in the protocol, so auditors can compare “why this segment” decisions without spelunking vendor-specific logs.

• Composable sustainability: Inject emissions or attention metrics as first-class filters in discovery; standardize the scoring rationale across platforms. (AdCP’s metadata scaffolding is built for this.)

Why AdCP is the agentic ecosystem play

AdCP is open and MCP-native, which aligns with where the broader industry is heading – major vendors and cloud providers are moving to standard, tool-centric agent interfaces, accelerating interop and security reviews. That lowers the cost of experimentation and increases the shelf-life of automations across vendors.

Governance: independent and repo-first

To move at agentic speed, it should stand up an independent, vendor-neutral foundation – think Prebid.org alike so no single company can steer the standard.

Principles:

• Neutral bylaws & elected steering across buy, sell, and tech participants.

• Repo-first development: spec, adapters, and reference agents in public repos.

• Open conformance tests for self-certification; no pay-to-play badges.

• Transparent versioning & security reviews via working groups.
  

Outcome: high velocity without capture, broad adoption without gatekeepers, and assurance for buyers and auditors through open, test-driven compatibility.

AdCP: Bottom line

OpenRTB standardized bidding; Prebid standardized selection; AdCP standardizes the agentic workflow.

If we want AI to be more than a dashboard copilot – and become an operational currency traded between platforms, AdCP’s MCP-based design and A2A compatibility are the right next step. The sooner brands get access to more working media, more transparency and the ability to port their brand values, definitions, safety guidelines systematically across their marketing plan.

(For deeper context, see my DMEXCO column on Context Agents – MCP is the practical backbone that lets those agents act across the ecosystem.)

Don’t miss any important news from the world of digital marketing, both nationally and internationally! Subscribe to the new DMEXCO newsletter “Digital Digest” and receive the latest information directly in your mailbox every Monday, Wednesday, and Friday.