{
 "acronym": "AdCP",
 "adoption": {
  "by": null,
  "level": "observed",
  "probe_run": null
 },
 "aliases": [],
 "analog": "Ad-tech capability and transaction API",
 "collisions": [],
 "discovery": {
  "mechanism": "Publisher-side well-known file (adagents.json) declares authorized agent endpoints and capabilities; brand-side well-known file (brand.json) declares operator identity, agent endpoints, and signing-key discovery; a hosted Registry API additionally resolves brands/publishers to AdCP agents by capability, category, or geography",
  "path": "/.well-known/adagents.json (publisher) and /.well-known/brand.json (brand/operator)"
 },
 "dispute": null,
 "evidence": [
  {
   "claim": "AdCP is a project of AgenticAdvertising.org and coordinates the full ad campaign lifecycle (discovery, accounts, product/creative, media buy execution, real-time serve-time matching via the Trusted Match Protocol, signals/audiences, governance, performance tracking, catalog sync, brand protection) across two transport protocols, MCP and A2A. Current documented version is 3.1.20. adagents.json and brand.json are the two well-known discovery files; a hosted Registry API also resolves agents by capability. Client SDKs are published and versioned for JavaScript/TypeScript (npm @adcp/sdk), Python (PyPI adcp), and Go (github.com/adcontextprotocol/adcp-go); an open-source community example implementation, the Prebid Sales Agent, is built by a Prebid working group (explicitly a community example, not a maintained reference implementation) with GAM integration. Foundation governance (structure, voting classes, Board composition, specification lifecycle, conduct rules) is documented in the project's CHARTER.md and published at agenticadvertising.org/governance.",
   "date": "2026",
   "label": "VERIFIED",
   "url": "https://docs.adcontextprotocol.org/"
  },
  {
   "claim": "AgenticAdvertising.org's own homepage (schema.org Organization markup) names itself 'the member organization pioneering a human-centric advertising future through Agentic AI,' founded 2025, and lists github.com/adcontextprotocol/adcp as its sameAs project repository, corroborating the docs site's governance claims from an independent page on the same organization's root domain.",
   "date": "2026",
   "label": "REPORTED",
   "url": "https://agenticadvertising.org"
  }
 ],
 "form": "well-known-file",
 "governance": "AgenticAdvertising.org, an industry organization of publishers, platforms, agencies, and technology providers; foundation governance (structure, voting classes, Board composition, spec lifecycle) is set out in the project's CHARTER.md and published at agenticadvertising.org/governance",
 "group": null,
 "history": [
  {
   "date": "2026-09-05",
   "note": "seeded from Exa Agent research run"
  },
  {
   "date": "2026-09-06",
   "note": "verified against docs.adcontextprotocol.org/dist/docs/3.1.20/intro"
  }
 ],
 "id": "adcp",
 "implementation": "shipping",
 "last_verified": "2026-09-06",
 "name": "Ad Context Protocol",
 "native_status": "Active open protocol, documented current release 3.1.20",
 "origin": {
  "date": "2025",
  "org": "AgenticAdvertising.org"
 },
 "part_of": null,
 "parties": [
  "agent-site",
  "agent-agent"
 ],
 "phase": "working-group",
 "probe": {
  "method": "GET",
  "on_absent": "404 or non-JSON response at /.well-known/adagents.json: the origin has not published AdCP publisher discovery (it may still run an AdCP agent without publishing the file, but that cannot be crawled)",
  "parse": "JSON body with `version`, `publisher.name`/`publisher.domain`, and an `agents[]` array of {url, protocol, capabilities}; a `get_adcp_capabilities` MCP/A2A call against a listed agent url confirms the endpoint actually speaks AdCP",
  "path": "{origin}/.well-known/adagents.json"
 },
 "question": "How can an AI agent discover ad inventory, execute a media buy, sync creative, and report performance across many ad platforms through one protocol?",
 "scope": "advertising",
 "spec_url": "https://docs.adcontextprotocol.org/dist/docs/3.1.20/intro",
 "step": "act",
 "summary_md": "The Ad Context Protocol (AdCP) is an open standard, published by the industry organization AgenticAdvertising.org, that lets AI agents discover ad inventory, set up commercial relationships, plan and execute media buys, generate and distribute creative, activate targeting data, run governance checks, and report performance across many ad platforms through one set of tasks instead of a bespoke integration per platform [1]. It explicitly targets more than programmatic exchange: it covers direct deals, sponsorships, broadcast, and out-of-home, since most global ad spend still flows through non-programmatic channels [1].\n\nAdCP solves ad-tech's fragmentation problem for agentic buying: a media buyer's agent sends one natural-language brief via `get_products` to every connected seller and receives back a common schema (pricing, forecasts, targeting, creative requirements) instead of comparing incompatible dashboards; `create_media_buy` then executes across sellers in one call, and `get_media_buy_delivery` aggregates reporting the same way [1]. Publisher-side discovery runs through a well-known `adagents.json` file (analogous to `robots.txt`) declaring authorized agent endpoints and capabilities, and brand-side identity/authorization runs through `brand.json`; a hosted Registry API additionally resolves publishers and brands to their AdCP agents by capability, category, or geography for broader discovery beyond known domains [1]. Trust is handled structurally: requests are signed with RFC 9421 HTTP Message Signatures against operator-published JWKS, idempotency keys make retried writes safe, and governance approvals travel as signed JWS tokens so no intermediate agent can forge a human sign-off [1].\n\nAdCP does not itself run ad auctions or hold inventory — it is a transaction and coordination layer that sits on top of two existing agent-transport protocols, MCP (for AI-assistant tool calls) and A2A (for agent-to-agent collaboration), using the same task schemas over either transport [1]. It also does not claim every listed agent's capability claims are true by construction: the protocol ships compliance storyboards (a testable, runnable verification suite) precisely because claims otherwise would not be independently checkable [1].\n\nAs of this verification, the documented current release is 3.1.20, with published, versioned client SDKs for JavaScript/TypeScript, Python, and Go [1]. The one open-source implementation surfaced in the docs, the Prebid Sales Agent (Python backend, TypeScript protocol layer, GAM integration), is explicitly labeled a community example built by a Prebid working group rather than a maintained reference implementation — a caveat AdCP's own documentation states outright, which this record treats as direct evidence of at least one independent, named implementer rather than as proof of broad adoption [1]. Foundation-level governance (voting classes, Board composition, spec lifecycle) is documented in a public CHARTER.md, which is more governance transparency than many single-vendor competitors in this space disclose [1][2]. What remains unverified: no independent count of publishers actually serving `adagents.json` in production, nor of live AdCP agent endpoints, was located in this pass — that would need this project's own probe run before adoption could be upgraded from `observed` to `measured`.",
 "track": "industry-body",
 "version": {
  "date": "2026",
  "label": "3.1.20"
 }
}