Picture a media buyer writing a plain brief - reach eco-conscious car buyers on connected TV in the US this week - and an AI agent taking that instruction and negotiating the campaign directly with a publisher's own systems, with no insertion order and no manual trafficking in between. The Ad Context Protocol is the shared language that makes that exchange possible. It is an attempt to give the machines now buying and selling advertising a common, open way to talk to each other, so the agentic era does not splinter into a hundred incompatible private integrations.
What is AdCP (the Ad Context Protocol)?
AdCP is an open-source communication protocol that lets AI agents - whether built by advertisers, publishers or ad tech intermediaries - interact using a common language. It standardises how those agents exchange structured information about audiences, inventory and campaign objectives, so a buyer's agent and a seller's agent can discover availability, agree terms and execute a deal without a human stitching the systems together by hand.
The protocol was published openly in late 2025 and is documented at adcontextprotocol.org. It is built on top of Anthropic's Model Context Protocol (MCP) and agent-to-agent (A2A) frameworks, which are the emerging plumbing for letting AI systems call tools and talk to one another. In plain terms, AdCP is the advertising-specific vocabulary layered on top of that general agent plumbing: MCP lets agents communicate at all, and AdCP tells them how to talk specifically about media.
How does AdCP work?
In an AdCP transaction, a buyer-side agent and a seller-side agent exchange structured messages describing what each side wants. The buyer agent expresses objectives - an audience, a channel, a budget, an outcome - and the seller agent responds with the inventory, terms and data it can offer. Because both sides share the same schema, the negotiation can resolve into a direct campaign that runs inside the publisher's own ad server or through a private or curated marketplace, rather than being auctioned impression by impression in a bidstream.
The design is deliberately asynchronous: a response can come back in seconds or in days, which leaves room for human-in-the-loop approval. That matters for publishers who do not want a machine signing off on inventory or pricing without editorial and brand-safety review. Advocates also argue the model is more auditable than today's programmatic plumbing, because agents transact directly with sellers and their ad servers as true direct buys, logging each step rather than passing budgets through layers of obfuscating intermediaries.
How is AdCP different from OpenRTB?
OpenRTB standardised real-time bidding - the impression-by-impression auction that powers most programmatic display today. AdCP standardises something different: the negotiation and coordination between agents that can happen before, around or instead of an auction. Scope3 chief executive Brian O'Kelley, one of the original architects of programmatic, compared its launch to the debut of header bidding, and the protocol is widely described as the OpenRTB of the AI era.
Crucially, AdCP is positioned as a parallel protocol to OpenRTB, not a replacement. The consortium behind it is explicit that publishers and platforms can run both at once - they are not mutually exclusive. A seller can keep its existing real-time bidding stack and add AdCP capabilities for agent-driven direct deals on top, adopting it gradually rather than ripping anything out. The intent is a new lane on the road, not a demolition of the existing one.
Who is behind AdCP?
The founding members include Optable, PubMatic, Scope3, Swivel, Triton Digital and Yahoo, with supporting members such as AccuWeather, Butler/Till, LG Ads, Raptive, Samba TV and The Weather Company. The group launched AdCP as an open-source initiative and said it is forming a non-profit governing entity so that no single company controls the protocol's evolution, with representation intended across publishers, advertisers, agencies and ad tech platforms.
Governance, more than the membership list, is the thing to watch. The history of advertising standards is full of well-intentioned protocols that stalled when large vendors moved to protect their own revenue. Whether AdCP stays neutral enough for rivals to trust it is the open question its own backers acknowledge, and the planned 2026 expansion of its scope to cover creative generation and performance attribution will test how broad that coalition really is.
What is agentic advertising, and why is it happening now?
Agentic advertising is the shift from humans operating advertising software to AI agents planning, negotiating, buying, optimising and reporting on campaigns largely on their own. It is happening now because the underlying agent technology has matured at the same moment that the old programmatic system is creaking under data silos, opaque fees and long, intermediated supply chains. eMarketer has argued that generative AI is taking over programmatic advertising in 2026, with agentic AI close behind, against a backdrop of US programmatic spend measured in the hundreds of billions of dollars.
AdCP is not the only answer to this moment. In January 2026 the IAB Tech Lab published its own Agentic Roadmap, which takes a different route: rather than a new standalone protocol, it extends established standards - OpenRTB, AdCOM, OpenDirect, VAST, the Deals API, Open Measurement and the privacy frameworks GPP and TCF - with agentic layers built on MCP, A2A and gRPC. Trade coverage framed the roadmap as the incumbent body keeping agentic buying anchored to existing standards, and notably as offering little explicit endorsement of AdCP. Publishers should expect a period of competing, overlapping approaches rather than a single settled standard, and separately should not confuse any of these with OpenAI's Agentic Commerce Protocol, which governs in-chat shopping and checkout rather than media buying.
What does agentic advertising mean for publishers?
For publishers, the promise of AdCP and the broader agentic shift is more direct, less intermediated selling. Instead of exposing inventory through multiple exchanges and waterfall setups, a publisher could present its inventory and contextual data straight to buyer agents, with transparent deal parameters and faster feedback. Agents could transact on audience segments, engagement or brand-lift outcomes rather than raw impressions, which opens the door to more flexible packaging and pricing. The risks are familiar: standards can be captured by the largest players, machine-speed dealing can outrun human oversight, and accountability for what an agent agrees to has to be designed in from the start.
But there is a structural limit worth naming clearly. Every one of these protocols - AdCP, the IAB roadmap, agentic commerce standards - assumes there is inventory to transact: an ad slot, a deal, a checkout. They standardise how machines buy and sell the advertising that already exists. They say nothing about the fastest-growing way an AI now uses a publisher's work, which is to read the page, extract the facts and compose an answer with no ad slot, no auction and no transaction at all.
Where agentic ad protocols stop, the AI answer begins
This is the layer blankspace operates in. When a Live Search Agent or LLM fetches a publisher's page to build an AI answer, there is no impression to bid on and no inventory for an agent to negotiate over - the value is extracted before any conventional ad transaction could occur. blankspace works at the CDN edge to detect that agent traffic as it actually arrives, and to inject contextual brand facts into the AI responses those agents generate, turning an otherwise uncompensated read into a monetisable, measurable event. Agentic ad protocols are important infrastructure for how machines will buy conventional media, and publishers should track AdCP and the IAB roadmap closely. But they govern the marketplace for inventory that exists, not the open-web answer surface where a growing share of a publisher's audience now lands and where, for now, no protocol pays for the read. Those are two different problems, and only one of them is solved by a buying standard.
Frequently asked questions
Is AdCP a replacement for OpenRTB?
No. AdCP is designed as a parallel protocol that complements OpenRTB rather than replacing it. OpenRTB standardises the real-time impression auction; AdCP standardises the agent-to-agent negotiation that can sit alongside it. Publishers and platforms can run both simultaneously and adopt AdCP capabilities gradually without removing their existing real-time bidding stacks.
Who controls AdCP?
AdCP launched as an open-source initiative with founding members including Optable, PubMatic, Scope3, Swivel, Triton Digital and Yahoo, plus supporting members such as AccuWeather, LG Ads, Raptive, Samba TV and The Weather Company. The group said it is establishing a non-profit governing entity with representation across publishers, advertisers, agencies and ad tech so that no single company controls the protocol. How neutral that governance proves in practice is still being tested.
How is AdCP different from the IAB Tech Lab's agentic roadmap?
They address the same moment by different means. AdCP is a new standalone protocol for agent-to-agent advertising transactions, built on MCP and A2A. The IAB Tech Lab's Agentic Roadmap, published in January 2026, instead extends existing standards such as OpenRTB, AdCOM, OpenDirect and VAST with agentic layers. They overlap, they are not formally aligned, and publishers should expect competing approaches before any single standard settles.
Is AdCP the same as OpenAI's Agentic Commerce Protocol?
No, and the two are easy to confuse. AdCP, the Ad Context Protocol, is about buying and selling advertising between agents. OpenAI's Agentic Commerce Protocol governs product discovery and instant checkout inside ChatGPT, so a user can buy a product without leaving the chat. One is advertising infrastructure; the other is shopping infrastructure. Both are part of the agentic shift, but they solve different problems.
Does AdCP help publishers get paid when an AI answer uses their content?
Not directly. AdCP standardises how advertising inventory is transacted between agents; it presumes there is an ad slot or deal to negotiate. When an AI assistant reads a publisher's page to compose an answer, there is no such inventory and nothing for an AdCP agent to buy. Capturing value from that uncompensated read is a separate problem, and it is the one blankspace addresses by detecting agent traffic at the CDN edge and injecting paid, contextual brand facts into the AI response itself.