AdCP 3.1 is GA: From MCP Access to Production-Grade Agentic Advertising
The first wave of agentic advertising was about access.
Across the industry, advertising companies and platforms have been standing up MCP servers, making their systems visible and callable by AI agents. That matters. It gives agents a way to discover inventory, retrieve reports, inspect campaign objects, and interact with systems that were previously locked behind human-operated interfaces or bespoke APIs.
But access is not the same as production infrastructure.
A basic MCP server can answer a query. It can expose a tool. It can wrap an existing API. That is useful, but it is still a 1.0 interface. It does not automatically solve authorization, lifecycle state, degraded delivery, reconciliation, catalog freshness, webhook observability, version compatibility, or the operational edge cases that appear once real budgets start moving through the system.
That is the shift AdCP 3.1 is built for.
The question is no longer just, "Can an agent call my platform?" The question is, "Can an agent safely plan, buy, deliver, monitor, and reconcile media through my platform in production?"
AdCP 3.1 is the beginning of that next phase: enterprise-class interfaces for agentic advertising.
The Problem AdCP Solves
Advertising has always depended on relationships, but the software layer has treated every relationship as a one-off integration.
Every buyer-seller connection tends to come with its own API behavior, inventory model, reporting format, approval process, reconciliation logic, and operational exceptions. A media team running campaigns across six channels is not just managing six channels. It is managing six different software problems.
Historically, the only way to scale past that has been to add headcount, or to ask teams to learn a new stack every time a client wants to try something the current one does not support. Scaling should come from adopting better technology, not from multiplying the number of people maintaining bilateral integrations.
That does not scale cleanly for humans, and it does not scale for agents either. An agent that has to speak a different operational language for every counterparty is not a breakthrough. It is complexity with a conversational interface on top.
AdCP exists to make the core advertising workflow interoperable: discovery, planning, negotiation, execution, reporting, and reconciliation. The goal is simple: implement the protocol once, then connect with conformant counterparties without rebuilding the workflow for every bilateral relationship.
In AdCP 3.0, that meant proving the workflow could run end to end.
In AdCP 3.1, it means hardening the protocol for production.
From Basic Agent Tools to Enterprise Interfaces
The industry is moving through a predictable maturity curve. First, companies expose tools. Then agents start using them. Then real workflows create real operational pressure. At that point, the interface has to mature.
That is where AdCP 3.1 comes in. A production-grade advertising interface needs to answer questions a basic tool endpoint often cannot:
- Is this agent actually authorized to act for this brand?
- What protocol version does this counterparty support?
- Has inventory changed since the last sync?
- Did a webhook fire, fail, retry, or get deduplicated?
- What happens when a creative, package, signal, or audience becomes unavailable mid-campaign?
- Can the buyer diagnose delivery problems without opening a support ticket?
- Can both sides reconcile what was planned, approved, delivered, and billed?
These are not edge cases once agents are running real buys. They are the operating requirements. AdCP 3.1 focuses on those requirements.
What Changes in AdCP 3.1
AdCP 3.1 does not replace the 3.0 model. It makes it more operationally complete. Every major addition addresses a point where agentic workflows can otherwise fall back into manual triage.
Silent failures get prevented. When a creative is pulled, an audience is suspended, or a package becomes unavailable mid-flight, the system needs to expose that state clearly rather than let it fail quietly. AdCP 3.1 makes degraded state queryable: agents can see which dependency is affected, which buys or packages are impacted, and what action is needed next.
Webhook activity becomes visible. A webhook that disappears into the void creates operational drag. AdCP 3.1 lets buyer agents check their own notifications directly — did it fire, when, and was it a duplicate — without having to ask the seller. Status, fire time, and idempotency keys are all visible, so delivery issues can be diagnosed without routing everything through seller support.
Catalog synchronization becomes scalable. Repeatedly re-fetching a full catalog just to check whether anything changed is wasteful. AdCP 3.1 adds feed mirroring with conditional fetch, so sellers can indicate when nothing has changed, notify partners when something has, and support targeted repair reads when gaps appear.
Brand authorization becomes verifiable. "Trust me, I represent this brand" is not sufficient for enterprise workflows. AdCP 3.1 requires brand verification responses to be signed, tightening the link between an agent's claim of authority and the brand's actual authorization.
Version compatibility becomes explicit. Production systems cannot rely on assumed compatibility. AdCP 3.1 introduces version negotiation, allowing sellers to advertise what they support and buyers to declare what they are built against. The handshake becomes explicit.
Together, these changes move AdCP from "the workflow can run" to "the workflow can be operated."
Upgrading from AdCP 3.0
AdCP 3.1 is additive — this is an upgrade, not a migration. No required 3.0 fields were removed. No existing 3.0-conformant client needs to stop working. Implementers can adopt the new surfaces on their own timeline.
For teams already running AdCP 3.0, the upgrade path is about maturity, not disruption. SDK authors, buyer agents, seller agents, creative agents, signal agents, and brand agents can each adopt the 3.1 capabilities that matter most to their workflow.
The full upgrade checklist is available in the AdCP 3.0 to 3.1 guide. The complete change list is in the 3.1.0 release notes, with an adopter-oriented walkthrough in What's New in AdCP 3.1.
Built by the People Running the Workflows
AdCP 3.1 did not come from abstract protocol design. It comes from implementation pressure: buyers, sellers, platforms, and engineers working through what happens when agents move beyond demos and start touching real media workflows.
That is how a protocol gets hardened. A basic integration surface can be designed in a room. A production interface has to be shaped by the people operating it.
The more real relationships run through AdCP, the stronger the protocol gets. Each implementation exposes what needs to be clearer, more observable, more verifiable, or more resilient. AdCP 3.1 is evidence that this feedback loop is working.
It also points to where the market is going.
MCP made advertising systems accessible to agents. AdCP defines how those agents transact through advertising systems in a way that can be trusted, audited, and scaled.
This Is the Next Phase of Agentic Advertising
Agentic advertising is moving from experimentation to infrastructure.
The first milestone was getting agents connected. The next milestone is making those connections production-ready.
That means interfaces that support real authorization, real delivery state, real reconciliation, real observability, and real compatibility across counterparties. It means moving beyond basic tool access into enterprise-class workflows.
That is what AdCP 3.1 is for.
AgenticAdvertising.org is the nonprofit member organization stewarding this work — over 150 members and 2,000+ practitioners across brands, agencies, publishers, technology platforms, and individual thought leaders — shaping how AI and advertising work together.
If you are building agentic advertising systems, now is the time to move from access to production readiness.
Read the AdCP 3.1 release notes. Review the upgrade guide. Start implementing the surfaces that make your agent interfaces enterprise-class.
And if you want to help shape what comes next, become a member of AgenticAdvertising.org.
Learn more: the AAO Academy's Level 1 course takes about 90 minutes and earns a digital credential. It's the fastest way to build working fluency in the vocabulary of the agentic future.