AdCP: The Open Standard for Agentic Advertising
Ad tech's blocker isn't whether agents work but that every platform speaks its own language. AdCP, built on Anthropic's MCP, standardizes three jobs — audience activation, curation, and media buy — so agents operate across stacks without custom glue code. The early wins are creative handling and signals discovery; media buying scales only with intermediaries and governance, because standards fix the plumbing, not the incentives.
Connect once. Operate anywhere.
Autonomous agents are no longer a lab toy in ad tech. They already run parts of live campaigns. The blocker is fragmentation: every platform speaks its own language. AdCP fixes this with an open, extensible standard for how intelligent agents talk to advertising platforms. It unifies three core jobs — Audience Activation, Curation, and Media Buy — so agents can plan and execute without custom builds.
Built on Anthropic’s Model Context Protocol (MCP). MCP handles auth, permissions, and context. AdCP uses that foundation so agents and platforms can exchange intent, options, and results in a clean, secure way. No rip-and-replace. Adopt it once; operate everywhere.
What AdCP enables
- One interface, many platforms. Connect once across your stack.
- Consistent intent. Express audiences, budgets, and outcomes the same way everywhere.
- Less glue code. Replace a patchwork of proprietary APIs with one spec.
- Uniform data. Inventory, pricing, performance, and delivery in a standard shape.
Audience Activation, Curation, and Media Buy are live today in the open-source reference implementation.
Where agents win first (Ari’s lens)
Ari Paparo’s take is practical and helpful: agents shine when they remove busywork and reduce error.
- Single-platform automation. If thousands of users type the same steps into your UI, agents will do it faster and better.
- Cross-platform with the same customer. Think creative + DSP, or data broker + DSP. Agents can pass precise requirements and payloads between parties and cut human rework.
Ari’s favorite early win: creative/asset handling. Clear schemas stop “AI guesswork.” Instead of “upload a banner,” the agent knows the exact asset type, size, format, and file limits. Fewer back-and-forths. Faster time to live.
Signals discovery is also promising: let agents query segment catalogs across providers, apply filters, and return a shortlist that fits the brief. Caveat (again from Ari): incentives and trust matter. Data sellers will market their segments; buyers need validation and governance. AdCP can standardize “what” and “how,” but the market still needs checks on “how good.” That check cannot be generated from inside the system: an agent can manufacture relevance, but not trust — trust has to be conferred from outside, by buyers, registries, and conformance regimes (What Agents Can’t Manufacture).
Media buying: be ambitious, but grounded
Ari is bullish on automation, cautious on agent-led transactions at scale. History shows tough issues in “programmatic direct”: rate cards, price discovery, and uneven data. Agents don’t erase those tensions. Two pragmatic paths forward:
- Intermediary-led execution. SSPs and cross-publisher networks can package supply and assume delivery risk. Think “curation by wire” instead of static deal IDs.
- Data-driven PG/PMP at scale. A buying agent with a tight audience definition can pre-negotiate across many publishers where the signal is common in an ad server or curation layer.
The takeaway: Media Buy via agents is viable, but the value likely accrues first to large publishers and intermediaries who can aggregate supply, signals, and accountability.
How it works (mirrors how teams think)
| Step | What happens |
|---|---|
| Intent | The agent declares objectives and constraints. |
| Discovery | Platforms respond with structured options for audiences, inventory, and pricing. |
| Activation | Approved plans launch using the Media Buy workflow. |
| Measurement | Results return in a consistent format for analysis and reporting. |
All exchanges are secure and context-aware per MCP.
Who uses AdCP
| Who | What they get |
|---|---|
| Advertisers & agencies | Access audiences and inventory directly. Plan on intent and outcomes, not line-item busywork. Get one view of pricing and performance across channels. |
| Ad tech platforms | Implement once to make exchanges, ad servers, and measurement tools agent-ready — while keeping your product edge. |
| Publishers & media owners | Make packages and pricing discoverable to agent buyers. Offer premium sponsorships and niche segments with less lift. Transact with transparency and control. |
Open, neutral, extensible
AdCP is open source and industry-led. A neutral consortium governs the standard with public docs, working-group review, and versioned releases. The core schema is extensible, with optional fields for proprietary data. Licensed under MIT to be easy to implement and hard to monopolize.
Founding Members: Optable, PubMatic, Scope3, Swivel, and Triton Digital
Supporting Members: AccuWeather, Butler/Till, Classify, Raptive, The Weather Company, Newton Research, Bidcliq, and Samba TV
Adoption path (start small, scale fast)
- Read the spec — GitHub
- Integrate one workflow (Audience Activation, Curation, or Media Buy) that maps to your product.
- Validate with conformance tests.
- Join the working group to guide releases, incentives, and guardrails.
Bottom line
- Agents reduce waste and speed execution.
- Clear schemas beat AI “best guesses.”
- Signals and creative win now; media buying scales with intermediaries and strong governance.
- Standards don’t fix incentives, but they make progress possible.
Ad Context Protocol (AdCP): the open standard for agentic advertising. Connect once. Operate anywhere. Learn more: adcontextprotocol.org