The Ad Context Protocol (AdCP) is an open standard, published and governed by AgenticAdvertising.org, that gives AI agents a common language to discover, plan, buy and report on media. In its first year it moved from a 2.0 launch on 15 October 2025 to version 3.2, which became generally available on 30 September 2026.

TL;DR. AdCP launched in October 2025 with more than 20 companies behind it and an initial scope of audience activation, curation and media buying. Version 3.0 in April 2026 widened it to brand identity, governance and creative, 3.1 in June hardened it for live operation, and 3.2 in September made proposals, holds and owed reports explicit. AAMP, IAB Tech Lab's Agentic Advertising Management Protocols, has developed alongside it, and AAMP 3.0 with OpenProposal is open for public comment until 10 November 2026.

Why does AdCP's release history matter to buyers and sellers?

A protocol that ships often changes what an agent can be asked to do, and an agent built against an earlier version may not use what the current one offers. The AdCP releases page on GitHub lists 116 tagged releases between 15 October 2025 and the end of October 2026, 72 of them stable and 44 betas or release candidates. That pace is normal for a young open standard, and it is why version pins and compatibility tables matter more than launch headlines.

The practical question for a buying or selling team is how to keep agents working while the standard underneath them moves. Alkimi is one worked example of a standard-neutral approach: it can work with agents built on AdCP or AAMP, with or without a DSP. Its Concourse benchmark tests any buying agent without asking which standard it uses, and its Gym lets buyers and sellers train their own agents and spar them against counterparty agents, regardless of protocol, before real money moves.

What has AdCP shipped since October 2025? A dated timeline

Each entry below comes from AgenticAdvertising.org's Versions and Compatibility page, its release notes and GitHub tags, or trade press published at the time.

15 October 2025: launch and version 2.0

AdCP launched in New York as an open standard for agent communication, backed by founding members Yahoo, Optable, PubMatic, Scope3, Swivel and Triton Digital plus a wider group of launch members, according to the launch announcement published by Samba TV. The initial release covered Audience Activation, Curation and Media Buy, and was built on Anthropic's Model Context Protocol (MCP). Version 2.0.0 was tagged the same day as the production release.

Digiday's launch explainer quoted a consortium spokesperson describing AdCP as "a parallel protocol to OpenRTB".

16 October 2025: the first agent-to-agent buy

The AdCP home page records that the first agent-to-agent media buy was executed the next day, with real money, inventory from LG Ads and AI agents handling creative, targeting and approval under human oversight. Adam Broitman, a partner at McKinsey & Company, is quoted there: "It felt like you were watching the first banner ad being served."

19 October and 22 November 2025: versions 2.1 and 2.5

Version 2.1.0 followed four days after launch, titled on GitHub as an asset schema simplification. Version 2.5.0, on 22 November, is described in the release notes as a developer experience and API refinement release. It added versioned schema paths so integrations could pin to an exact version.

January 2026: the last 2.x release and the first 3.0 beta

Version 2.5.3, the last release in the 2.x line, shipped in January 2026. The 3.0.0-beta.1 tag followed on 26 January, and development moved to the 3.x line from there.

21 and 22 April 2026: AdCP 3.0

AgenticAdvertising.org announced AdCP 3.0 on 21 April 2026, six months after launch, and the versions table lists 3.0.0 as generally available on 22 April. The What's new in AdCP 3.0 page lists a Brand Protocol, governance, an Accounts Protocol, media planning through proposals with budget allocations and delivery forecasts, and a channel model that grew from 9 to 20 channels. The announcement framed governance as "Embedded Human Judgment", in which humans stay in the loop on decisions with real consequences rather than on every operational step.

18 June 2026: AdCP 3.1

Version 3.1.0 was a minor release that the What's new in AdCP 3.1 page calls "the production-hardening release", with every change additive over 3.0. It let buyer agents see why a buy was impaired, inspect their own webhook deliveries and tie optimisation goals to vendor-attested measurement.

1 August 2026: version 2 reaches end of life

The v2 sunset page records that the 2.x line stopped receiving patches on 1 August 2026. AgenticAdvertising.org certification and verified status in its registry now require version 3, and the 2.x schema URLs still resolve so existing integrations do not break silently.

30 September 2026: AdCP 3.2, released as 3.2.1

AdCP 3.2 became generally available on 30 September 2026, with 3.2.1 as the first stable release because the 3.2.0 number was never released. The What's new in AdCP 3.2 page sums it up: "3.1 made agents safe to operate. 3.2 makes the deal between them explicit."

The release introduced seven task-specific tools for running a buy, including request_proposals, buy_products and accept_proposal, with MCP schemas 71 to 94% smaller than the 3.1 equivalents. It also added shared budgets with buyer cost controls and an experimental Reliable Reporting pipeline.

22 October 2026: the 3.3 beta opens

The 3.3.0-beta.0 tag was published on 22 October 2026, and the documentation banner describes the beta as bringing "scheduled, human-gated, and offline media in the same buy as programmatic". Patch releases on the 3.2 line have continued alongside it, with 3.2.8 tagged on 28 October.

What does a year of AdCP change for buyers?

For buyers, the biggest change is that the proposal stage now has a formal shape. Under 3.2 a buyer agent can move from a brief to seller-authored plans, negotiate typed terms, hold inventory and accept one exact snapshot, according to the 3.2 overview. That is the territory of RFP chasing and IO paperwork, which media teams have long handled through email, calls and spreadsheets.

The same overview says reports the seller owes "can no longer silently disappear", because the reporting pipeline is now part of the protocol, though still experimental. For buyers, that gives reconciliation and make-good conversations a shared starting point: what was owed, and when.

Approval stays with people. When a governance agent is configured, AdCP governance escalates actions above a pre-approved threshold to a human. At launch, Scope3 co-founder Anne Coghlan told AdExchanger: "Where we are right now, there is always a human in the loop."

What does it change for sellers?

For sellers, 3.2 is about selling the way their own platform works. The 3.2 overview says a seller can optimise a shared budget natively against the buyer's cost controls, state what a product needs before it can run and declare how audience data reaches it, including through clean rooms.

Version 3.0 had already given publishers and broadcasters a model for programming, with collections, instalments and break-based inventory.

The operational discipline is version honesty. The 3.2 guidance tells sellers to advertise "3.2" only when they serve it and to declare only the capabilities they implement, which is what lets a buyer agent trust the answer it gets.

Where does AAMP fit alongside AdCP?

AAMP, IAB Tech Lab's Agentic Advertising Management Protocols, is a separate framework developed under IAB Tech Lab. On 22 September 2026 IAB Tech Lab announced AAMP 3.0, introducing OpenProposal, a specification for how sellers represent products and respond to buyer briefs in an agent-readable format. OpenProposal connects to AdCOM, OpenDirect and the Deals API, and IAB Tech Lab extended its public comment period to 10 November 2026, as IAB Tech Lab reported.

AAMP and AdCP now both address the step between a buyer's brief and a buy commitment: AAMP through OpenProposal, AdCP through tasks such as request_proposals and accept_proposal. Buyers and sellers are likely to meet agents built on either, so the useful question is less which standard an agent speaks and more whether it behaves well on that standard. For a side-by-side map, see AAMP, AdCP, A2A and MCP: which protocol does what.

What should teams do with a standard that ships this often?

Treat the protocol version as something to test, not a fixed fact. Pin versions, check what a counterparty lists in supported_versions, and rehearse upgrades before they touch live budgets. The division of labour the protocol now encodes is clear: agents take on the RFP chasing, IO paperwork, reconciliation and make-good disputes people do not enjoy, and people keep the judgement and approve the deals that matter.

Alkimi builds for that split through earned autonomy, in which an agent moves from Observe to Recommend, Draft, Human-approved action and Bounded automatic action as it proves itself. Its Marketplace is neutral across standards, keeps approval rights with people at the deal level and records each agreement in a DealSheet held by both parties. In Concourse Bench v1, 55 of 96 runs completed a buy, which is why testing an agent before it trades matters whichever standard it speaks.

Frequently asked questions

What is the Ad Context Protocol?

AdCP is an open protocol, published and governed by AgenticAdvertising.org, that lets AI agents discover, plan, buy, sell and measure media across channels. It launched on 15 October 2025 and was built on Anthropic's Model Context Protocol.

What is the current version of AdCP?

Version 3.2 has been the current production release since 30 September 2026, shipped as release 3.2.1 with the wire pin "3.2". Versions 3.1 and 3.0 are still maintained, and a 3.3 beta opened on 22 October 2026.

Is AdCP version 2 still supported?

No, the 2.x line reached end of life on 1 August 2026 and receives no further patches. AgenticAdvertising.org certification and verified registry status require version 3.

Does AdCP replace OpenRTB?

At launch the AdCP consortium described it as a parallel protocol to OpenRTB and said publishers and platforms can run both at once. AdCP covers the agent conversation around a deal, while OpenRTB continues to handle real-time bidding.

How do AdCP and AAMP relate?

They are separate standards from different bodies: AdCP from AgenticAdvertising.org and AAMP from IAB Tech Lab. Both now cover the move from a buyer's brief to a buy commitment, and teams are likely to work with agents built on either.

AdCP One Year On: What the Ad Context Protocol Has Shipped Since October 2025