Skip to main content
The L0 wire layer is JSON-over-HTTP framed by published JSON Schemas. This page is the reference for getting the schemas — where they live, how to pin a version, how to verify supply-chain provenance, and the directory shape inside a release. If you’re picking an SDK rather than the schemas themselves, see Choose your SDK.

Schema access

AdCP schemas are available from two sources: Both sources contain identical schemas. The GitHub repository includes all released versions with bundled schemas committed directly to the codebase.

Schema identity and offline resolution

The latest tree and releases cut after this behavior was introduced use canonical HTTPS URIs for root $id values and external $ref values, for example https://adcontextprotocol.org/schemas/{version}/core/product.json. These URIs are stable schema identifiers; a validator does not have to fetch them over the network. Older release directories are immutable and retain their original /schemas/... references; for offline use of those releases, select their bundled/ schemas. When using the modular schema tree from a downloaded tarball, configure the validator or code generator to map the canonical https://adcontextprotocol.org/schemas/{version}/ prefix to the extracted schemas/ directory. Tools that cannot register URI-to-file mappings should use the corresponding schemas/bundled/ artifact, which has external $ref values resolved inline. For Python’s jsonschema, the referencing registry can perform that mapping without network access:

One-shot protocol bundle

Syncing hundreds of individual schema files adds up. Every AdCP release also publishes a single gzipped tarball containing the complete protocol — schemas, compliance storyboards, and the OpenAPI registry — so clients can pull one artifact instead of crawling the tree. Every tarball extracts into a single adcp-{version}/ directory (safe extraction, no tarbomb). Inside:
Verify the checksum before extracting:
Pull it once per version, cache by SHA, and you have everything needed to validate requests, run storyboards, and render documentation offline. The @adcp/sdk sync-schemas command uses this under the hood. Available tarballs are also listed at /protocol/.

Verifying protocol bundle signatures

The SHA-256 sidecar lives on the same origin as the tarball, so it only protects against in-transit tampering. For supply-chain protection — proving the bundle came from the AdCP release workflow and was not swapped for a malicious one even if the host were compromised — every released {version}.tgz is also published with a Sigstore detached signature. The signature is produced by the GitHub Actions release workflow using keyless OIDC: there is no long-lived AdCP signing key to leak. The certificate binds the signature to the workflow identity that issued it.
cosign verify-blob exits non-zero if the signature was made by anything other than the AdCP release workflow, even if the SHA matches and TLS is valid. Use this in any pipeline that ingests the protocol bundle as an enforcement source. The @adcp/sdk, adcp-client-python, and adcp-go SDKs perform this verification automatically when the sidecars are present. The refs/(heads|tags)/.* wildcard is intentional — releases sign during the push-triggered workflow run, so the cert subject names the release branch (e.g. refs/heads/3.0.x for v3.0.1+, refs/heads/main for v3.0.0). The trust gate is upstream release.yml’s on.push.branches allowlist, not the consumer’s regex. Literal-allowlist regexes ((main|2\.6\.x)-style) silently break every time a new maintenance branch is added — see Verifying protocol tarballs for the full trust model and the cert-subject-by-release lookup. Older releases that predate signing, and versions republished out of band (bypassing the signing workflow), remain checksum-only — clients should treat missing sidecars as a “checksum-only” trust level rather than a verification failure.

Compliance storyboards

Storyboards live alongside schemas at /compliance/{version}/. They define the test scenarios AAO runs to verify an agent’s capability claims. Declare supported_protocols (for protocol baselines) and specialisms (for narrow capability claims) in get_adcp_capabilities — the compliance runner executes the matching bundles to verify. See the full Compliance Catalog for every protocol and specialism an agent can claim.

Common schemas

For AI coding agents: point your coding agent to https://docs.adcontextprotocol.org/mcp for MCP integration documentation.

Schema versioning

AdCP uses semantic versioning. Choose the right path for your use case: The same version semantics apply to /schemas, /compliance, and /protocol/{version}.tgz — one release cuts all three. Pin to an exact version for stability:

Development

Use the major version alias to stay current with backward-compatible updates:

SDK type generation

Bundled schemas

For tools that don’t support $ref resolution, use bundled schemas with all references resolved inline. Bundled schemas are available from both the website and GitHub:

MCP 2026 tool-schema projection

Beginning with AdCP 3.2, every release also includes a self-contained JSON Schema 2020-12 projection for MCP 2026-07-28 tool discovery:
The projection manifest maps every AdCP tool to downloadable inputSchema and outputSchema files. Servers can select and embed these schemas when building their tools/list response; publishing the artifacts does not automatically change a server’s MCP registration. MCP disables automatic network dereferencing by default, so each projected schema is a compact, self-contained document containing only local fragment references. The 3.2 projection changes schema syntax, not the AdCP payload contract. It is generated from the canonical draft-07 schemas and MUST preserve the same validation outcomes. In particular, it does not add unevaluatedProperties: false or otherwise close extension points. The build tests inspect every projected schema for dialect, local-reference integrity, and self-containment; enforce AdCP-defined defensive depth, schema-object, and compact-serialization byte bounds following MCP guidance; compile representative schemas from every protocol; and compare the source and projected dialects across representative instances. Repeated shared schemas are stored once under $defs rather than recursively inlined.

Production surface profile

The MCP projection also publishes a filtered production profile:
For AdCP 3.2, this is the clean active structural catalog: it excludes the compliance-only controller and tools deprecated by 3.2, including the get_products and list_creative_formats compatibility facades. Its schemas remove only presentation annotations (description, enumDescriptions, title, examples, and $comment), so validation semantics are identical to the full MCP projection. Each manifest tool retains its protocol classification for deterministic selection. The profile is not a recommendation to load all active AdCP tools into one agent context. It currently spans 66 tools across the protocol families. A production host MUST expose only the protocols and tools it implements and SHOULD further select the smallest capability-appropriate subset for each agent session. Use the full projection for documentation, compatibility, and conformance; use this profile as the filtered catalog and structural validation source from which a host builds that subset. Removing descriptions makes the artifacts smaller, but does not by itself solve tools/list context cost.

Active role catalogs

Hosts and clients can use one of two role-filtered catalogs instead of selecting active tools from the entire production profile:
The media-buy catalog covers active 3.2 operations for a seller-hosted product-to-delivery role. It includes product discovery, proposals, purchase, control, reporting, audiences, catalogs, event sources, accounts, governance, and separately synchronized creatives. These operations are one production lifecycle rather than separate “sales” and “sales lifecycle” surfaces. Creative construction is deliberately excluded. The creative catalog covers creative construction, transformation, preview, validation, catalog inputs, library synchronization, delivery, accounts, governance, usage, and task management. Shared account, catalog, and creative-trafficking tools intentionally appear in both catalogs; they describe role-oriented active surfaces, not mutually exclusive protocol ownership. These are active-3.2 catalogs, not complete 3.x server registrations. A server that supports callers using the deprecated 3.x compatibility facades must also advertise the applicable get_products, create_media_buy, and update_media_buy definitions from the full projection. Cross-agent buyer orchestration can additionally require signals, brand, external governance, property, or content-standard services that are intentionally outside the seller-hosted role catalog. Clients that combine compatibility facades with an active role catalog use the full projection’s task_result_resolution, because the role-scoped resolver intentionally covers only task types present in that active catalog. Each role also publishes an input-only client prompt view:
Model-context manifests contain only structural inputSchema entries. They are client-side prompt projections, not standalone MCP tools/list registrations: servers SHOULD continue advertising outputSchema, and clients SHOULD validate structured results with the parent role catalog. A controlled client can omit the output schema only when assembling its model prompt while retaining the parent response schemas and task-result resolution metadata out of band. Structural schemas deliberately omit descriptions. To support tool selection, clients combine the model-context inputs with each live MCP tool’s concise name and description; the downloadable model-context manifest is not a description catalog by itself. Every tool in the published active Media Buy and Creative role catalogs carries a concise manifest summary that a host can use as that live description.

Capability-selected runtime projection

AdCP 3.2 hosts MUST derive a live MCP tool surface from the release manifest and the tools the endpoint can actually dispatch. A runtime projection is a selection of the generated per-tool bundles, not another schema profile and not a second hand-maintained tool catalog. The deterministic selection algorithm is:
  1. Build implemented_tools from the endpoint’s dispatch registry. Every name MUST exist in the canonical release manifest. Do not infer implementation from documentation, a role profile, or a protocol claim.
  2. If the session has no narrower capability scope, select implemented_tools. Otherwise, select the implemented tools whose manifest protocol is enabled, unioned with exact enabled tool names. Convert supported_protocols snake case to manifest kebab case (media_buymedia-buy) before comparing. Exact tool claims that are not implemented are configuration errors; hosts MUST fail closed rather than advertise them.
  3. Treat protocol as ownership metadata, not dependency closure. Shared task, account, or discovery tools are included only when the host adds their exact names. Selection never pulls in neighboring tools implicitly.
  4. Production projections MUST exclude the compliance protocol. Deprecated compatibility facades are included only when the endpoint really implements and advertises them; deprecation alone is not a runtime filter.
  5. Sort selected names lexicographically. For each selected name, emit one MCP tools/list entry containing name, the optional manifest summary as the live description, and the corresponding self-contained inputSchema from the MCP projection. Do not emit unselected tools or load response schemas into the model-facing list.
  6. Keep the release manifest and response bundles available outside model context. SDKs validate a direct result through that tool’s response_schema; they resolve terminal polling results through task_result_resolution.
The coarse protocol set normally comes from supported_protocols. Exact names come from the active capability blocks (for example lifecycle_tools, repair_tasks, and projection_tasks) plus shared tools the host exposes for that session. The implementation registry remains the upper bound in every case. Unknown protocol or tool names, duplicate selector inputs, and production attempts to expose compliance tools are errors. For caller-side version adaptation, the live discovery result is authoritative. An SDK calls a current split 3.2 tool when that name is present. If it is absent, the SDK checks the current tool’s legacy_fallback manifest entry and may use the named legacy tool only when that name is live: direct is a one-call translation, orchestrated must preserve the current operation’s state and idempotency semantics across the sequence, and none is unsupported. SDKs MUST NOT infer a fallback from a shared protocol classification. The checked-in production, media-buy, and creative profiles remain useful catalogs and role-oriented starting points. They are not mandatory runtime combinations. The complete canonical projection remains the authority for documentation, conformance, compatibility analysis, code generation, and lazy response validation. AdCP 4.0 will make JSON Schema 2020-12 the canonical source dialect. That major-version migration is where the protocol may selectively use unevaluatedProperties, dependentRequired, dependentSchemas, and other 2020-12 semantics to tighten contracts. Published v3 schemas remain draft-07 for their support lifetime.

Website access

GitHub access

Bundled schemas are committed to the repository at dist/schemas/{VERSION}/bundled/:

Directory structure

index.json is the directory authority. It declares the bundle’s published_version, stability metadata, and protocol_layers: the negotiation layer (media-buy, creative, signals, account, governance, brand, sponsored-intelligence) and the decisioning/serving layer (trusted-match).

Bundled schema categories

All request/response task schemas are bundled: See the schema registry for all available schemas.

Version discovery

Do not infer the canonical version from directory order or versions[0]; pre-release artifacts remain discoverable for pinned historical builds. Use latest_stable or the aliases map for canonical stable selection. Check Release Notes for version history and migration guides.

Registry API

The AgenticAdvertising.org registry provides a public REST API for brand resolution, property resolution, agent discovery, and authorization validation. No authentication required.

Registry API Reference

Resolve brands, discover agents, and validate authorization via REST.