Two new acronyms crashed our calendars this month. At Base.Camp Lisbon, Scope3’s Alex Oakden introduced the Ad Context Protocol (AdCP). Two days later, at the IAB Tech Lab International Summit, Shailley Singh unveiled the User Context Protocol (UCP).

Here, we distill what each protocol is trying to standardize, and report back on how publishers in the room reacted when the slides ended and the real questions began.

AdCP: agents, a common language, and a “storefront” for sellers

It starts with agentic AI: a model, plus tools, plus a reasoning loop that can take a marketing brief and turn it into actions across systems. The recent Model Context Protocol (MCP) gave agents a common way to call tools, but AdCP layers an advertising-specific language on top. A buying agent and a selling agent can negotiate, then push approved campaigns into ad servers. Think of it as MCP for media, not auctions.

“AdCP is not a Scope3 product. We are building toward a standardized language,” Alex said, emphasizing a multi-company effort that includes buy- and sell-side platforms, publishers, and intermediaries.

AdCP comprises a Media Buy Protocol that governs the buyer–seller agent conversation, a Signal Activation Protocol that enables AI agents to discover, activate, and manage various data signals (including first-party data for audience targeting), and a Creative Protocol that helps generate AI-powered creatives that align with platform requirements and scales asset creation across various formats. Once a human approves the plan, the agent can push to your ad server of choice and return normal ad-serving reports.

The publisher pitch is simple: “Think of AdCP as your digital storefront to evangelize your unique assets to brand agents.”. AdCP aims to combine direct’s transparency and creative latitude with programmatic’s scale, again, without forcing everything through an auction.

Protecting publisher equity in the age of agents

If there was a chorus line in Lisbon, it sounded like this: ‘Great idea. Now show me measurement, governance, and how you avoid the next fraud cycle.’ Multiple publishers pressed on whether outcome trading means buyers must reveal targets, how guardrails will be enforced, and how pricing floors won’t be hammered by automated price-discovery pings.

In response, Alex pointed to a governance working group, activation-layer controls, and the fact that agentic deals look more like direct than open auction, which should limit the worst behavior.

Another concern was déjà vu from OpenRTB – a standard that flattens differentiation. Alex argued that sellers can encode what is unique about their inventory right into the agent’s merchandising prompts, so you appear to buyers more like a well-written Airbnb listing than a generic line item.

UCP: signals, embeddings, and translating human briefs

Meanwhile, UCP is an open standard that defines how intelligent ad agents exchange three specific signals: user (in privacy-safe form), context (page, app, environment), and reinforcement (propensity or inferred intent from observed behavior).

As Shailey explains, “the idea behind UCP is to capture those signals and put them into an agent as a negotiation between the seller agent and the buyer agent for collaborating - for understanding the data before you get to the activation.”

What’s more, unlike traditional text-based exchanges which are too slow for modern programmatic advertising, UCP leverages ‘embeddings’, which are compact, learned vector representations that efficiently encode complex signals in a privacy-preserving, interoperable format. For example, “sports fans” could be blended with “Rolex fans,” then the audiences weighted appropriately. He also stated that the objective is to make UCP fast enough at the exchange so agents can eventually resolve the signal schema at bid time, even if they’re not able to do this now.

Where does UCP actually live?

The most persistent question from publishers was architectural. Should UCP anchor in the Order Management System? Several publishers argued yes, because the OMS is the system of record for products, pricing, and finance. Others warned many OMSs are not yet technical enough, and nobody wants to be single-stacked into an ad server. The pragmatic take? Proofs may start via GAM integrations, but the end-state needs a neutral activation layer that speaks OMS, SSP, and whatever comes next.

Guardrails were the second concern. Publishers want up-front limits that curb spammy price-discovery, or multiple agents all switching to learning mode simultaneously and probing their inventory at once. This explosion in request volume could lower dynamic pricing floors and increase compute cost. So their ask was to cap abusive request patterns, codify limits, and resist reseller-style middlemen. As Rob himself quipped, “Some claim the future is frictionless, but life has friction, and some of it is valuable.” In other words, design useful friction, not useless fees.

Make the juice worth the squeeze

Of course, the question on most publishers’ lips was: which context protocol will become the default? Shailley was explicit that “UCP isn’t competing with AdCP; it’s complementary”.

In our Base.Camp wrap session, the refrain about new context protocols was value exchange. Publishers wanted reassurance that, if they added new steps, the steps must add measurable value, and the value must be shared. They asked for a real feedback loop on outcomes, not a vague promise to optimize. There were concrete worries about agent-to-agent negotiation gaming the pricing floors, a renewed push for price clarity over traditional floors, and a reminder that electricity and compute costs will ride shotgun with any plan to answer more briefs more quickly.

There were discussions around collective strategies for publishers, such as regaining control over attribution and inventory to reduce reliance on intermediaries who profit from it. There was an emphasis on using available tech for self-curation and deciding which signals to provide based on pricing tiers, following a ‘you get what you pay for’ model.

There was also a concern about the talent pipeline. If AI agents hollow out all the junior work, how will humans train for senior roles? The takeaway was to pair efficiency with skills, i.e., build the feedback loops we do not have, get teams into the working groups, and make shared learning part of the plan, not an afterthought.

On the plus side, if agents expose more of the brief than today’s bidstream, a capable seller agent should help you see why you did or did not win, then merchandise accordingly. If an activation layer truly reduces duplicate middle-layers, publishers could claw back both transparency and margin. But there’s also a clear flip-side: if we repeat early programmatic mistakes, we’ll get early programmatic outcomes. So choose your adventure wisely.

AdCP gives buyers and sellers a way to talk using something richer than a bid request, and UCP gives agents a common syntax for signals, privacy, and intent. Together, they have the potential to make direct selling feel scalable, and programmatic feel less generic. But that’s providing the industry designs the guardrails it wishes it had last time. Build the storefront, decide what not to stock, and price like you mean it. Then resist the cult of frictionlessness, because some friction accumulates value, and value is what keeps a protocol from becoming yet another initialism we all regret learning.