• Content Metadata Marketplace Supply Specification (MVP)
• Purpose
  • In scope
  • Out of scope
• AI System Request for Content
  • Object: AISystem
  • Object: AISystemUse
• Content Owner or Marketplace Response
  • Object: Package
  • Object: Scope
  • Object: Text
  • Object: Video
  • Object: Image
  • Object: Audio
  • Object: Retrieval
• Implementation Guidance
  • Function Definitions
  • Sub-Function Definitions
  • Access Format Type Definitions
  • JSON Examples
    • Example 1: Unauthorized Request
    • Example 2: Authorized Request
    • Example 3: Curated Selection of Multiple Asset Types
    • Example 4: Image Creation
    • Example 5: Podcast Full Feed Authorized
    • Example 6: Podcast Published After a Date
    • Example 7: Video Creation
    • Example 8: Agent Actions

Define the object model and transport format for describing content assets offered by Inventory Owners and/or Content Marketplace to an AI System. It aims to build the foundation for a marketplace that supports Content Owners while providing enough flexibility to future proof for new use cases.

This API will standardize a language that allows inventory owners to communicate metadata about the content on offer, where the terms to use that content are, and how to access the content. It also allows AI Systems to tell Content owners and/or Marketplaces the action they would like to perform using the content package

• Bot Blocking - while Content Owners are strongly encouraged to block crawling until and unless some commercial agreement is in place, Content Owners must work with their partners to block crawling from AI Systems at the level they deem appropriate for their business
• Licensing Terms - Commercial terms between Content Owners, Marketplaces, and AI Systems must be negotiated a priori to this API
• Clearing House Functionality - This API can be used to communicate where AI Systems can find terms to license and access content, but token issuance, counting, and payment are out of scope.
• Supply Discovery - It is assumed that AI Systems have an understanding of the content they’d like to license. Where discovery is required, AI Systems should work directly with Marketplaces and Content Owners.
• Reporting - The CoMP API is a communication protocol only, as such, it does not explicitly support reporting, but it is strongly recommended that the Content Owner receives reporting from the AI System to track and monitor usage against terms laid out in the license.

Information about the AI System making the request to scrape the content

Information about the Content being requested, and how the AI system will utilize the content after it has been crawled

High-level information about the package of inventory to be included in the crawl by the AI System

Information about the scope of the elements included in the package

Information about each text-based asset included in the package. One object per individual text-based asset should be sent.

Information about each video asset included in the package. One object per individual video asset should be sent.

Information about each image asset included in the package. One object per individual image asset should be sent.

Information about each audio asset included in the package. One object per individual audio asset should be sent.

Information for the AI System to access the content once a token has been issued

The AI System identifies itself and its intent but indicates it does not yet have an issued token.

Request:

{
  "aisystem": {
    "name": "open-learner-ai.org",
    "ua": "OpenLearnerBot/2.1",
    "id": "TL-REG-445566",
    "aisysuse": {
      "lid": "PENDING_REVIEW",
      "aiauth": 0,
      "uri": [
        "https://primary-publisher.com"
      ],
      "scope": 0,
      "function": [2],
      "subfn": [0],
      "resdis": 0
    }
  }
}

Data Points from the Specification

• aisystem.name: The canonical domain of the AI System requesting access.
• aisysuse.lid: Since a token has not been issued, the system should review terms in package.licenseurl.
• aiauth (0): The AI system plans to authenticate via UserAgent because it lacks a specific token or IP range permissioning.
• scope (0): Indicates a request for the full corpus (e.g., the entire website).
• function (2): Explicitly states the function is ai-train.
• subfn (0): Further specifies the sub-function as training.
• resdis (0): Confirms that results will not be displayed to a human user.

Relevant Implementation Notes

• Bot Blocking: The specification notes that Content Owners are strongly encouraged to block crawling until a commercial agreement is in place.
• Licensing: Commercial terms must be negotiated a priori to the use of this API.
• Clearing House: While this API communicates intent, it does not handle the actual payment or token issuance.

In this scenario, the Content Owner responds to the unauthorized request by providing a Package object that redirects the AI System to the required licensing terms. The response specifies that no authorization is currently granted (auth 0), but points the requester to a licenseurl to negotiate terms a priori to accessing the full corpus.

Content Owner Response: Redirect to Licensing

{
  "package": {
    "id": "PKG-REF-99-PENDING",
    "title": "Full Corpus Access - Licensing Required",
    "seller": "primary-publisher.com",
    "licenseurl": "https://primary-publisher.com/licensing/ai-terms",
    "citation": 1,
    "scope": {
      "scope": 0,
      "max": 1,
      "ctype": [4]
    },
    "retrieval": {
      "auth": 0,
      "endpoint": "https://api.primary-publisher.com/v1/unauthorized",
      "type": [2]
    }
  }
}

Key Object Details from the Specification

• package.licenseurl: Used to communicate where the AI System can find terms to license and access content, as token issuance and payment are out of scope for this API.
• package.citation (1): Indicates that citation of the Content Owner is required if the content is eventually used.
• scope.scope (0): Confirms the response acknowledges a request for the full corpus.
• scope.max (1): Indicates there is an upper limit on how much the AI system may crawl, and the system should reference the license to determine that limit.
• scope.ctype (4): Denotes that "all" content types (text, video, image, audio) are included in this corpus definition.
• retrieval.auth (0): Specifies that currently "none" (no authorization) is active for this requester at this endpoint.
• retrieval.type (2): Informs the AI system that content is typically delivered via API once authorized.

The AI System has a pre-existing commercial agreement (indicated by a valid License ID) and specifies that it will authenticate using its IP range, a common method for server-to-server crawling permissions. The AI System global-news-aggregator.com is requesting access to a specific section of content for grounding and RAG (Retrieval-Augmented Generation) purposes.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "global-news-aggregator.com",
    "ua": "GlobalNewsBot/5.0",
    "id": "TL-REG-882211",
    "aisysuse": {
      "lid": "LIC-998-ALPHA-2026",
      "aiauth": 1,
      "uri": [
        "https://fin-times.com/markets/commodities"
      ],
      "scope": 1,
      "function": [3],
      "subfn": [1, 2],
      "resdis": 1
    }
  }
}

Breakdown of Metadata

• aiauth (1): The AI system explicitly states it plans to authenticate via IP range.
• lid: A unique license reference is provided, assuming the system has a license to perform the function a priori.
• scope (1): The request is limited to a specific section of the site.
• function (3): The intent is ai-input, which includes RAG and grounding.
• subfn (1, 2): Specifically defines the use cases as RAG and grounding.
• resdis (1): Indicates that the results will be displayed to a human user.

Content Owner Response (Package)

The Content Owner acknowledges the License ID and confirms that the content is available for retrieval via a NewsML feed, which is structured for machine-readable distribution.

{
  "package": {
    "id": "PKG-FIN-COMM-001",
    "title": "Commodities Market Deep-Dive",
    "seller": "fin-times.com",
    "licenseurl": "https://fin-times.com/legal/terms/LIC-998",
    "citation": 1,
    "scope": {
      "scope": 1,
      "max": 0,
      "ctype": [0],
      "text": [
        {
          "title": "Gold Prices Surge Amid Volatility",
          "wordcount": [1200],
          "pubdate": "2026-02-26T14:30:00Z",
          "author": ["Alex Sterling"],
          "sourcetype": 0,
          "provenance": 1,
          "provent": "c2pa.org"
        }
      ]
    },
    "retrieval": {
      "auth": 0,
      "endpoint": "https://newsml.fin-times.com/v3/commodities",
      "type": [6]
    }
  }
}

Key Response Elements

• max (0): The AI system has unlimited access within this scope based on the license.
• citation (1): Citation is required, which aligns with the AI system's intent to display results to humans (resdis: 1).
• retrieval.auth (0): Listed as "none" because the authentication is handled at the network level via the IP range specified in the request, rather than an API key or OAuth.
• retrieval.type (6): Delivery is via NewsML, a structured format that does not require human readability for the initial transfer.

An AI System is looking for specific visual content to help an "Agent" perform tasks (e.g., identifying products or locations) based on a curated list provided by a Marketplace. The AI System requests a specific Curated Selection of high-quality multimedia assets. The Content Owner responds with a package containing a specific Image and Video asset, providing detailed metadata for each to ensure the AI System understands the source and length of the content.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "vision-agent.ai",
    "ua": "VisionAgentBot/1.2",
    "aisysuse": {
      "lid": "LIC-CURATED-777",
      "aiauth": 2,
      "scope": 5,
      "function": [1],
      "subfn": [4],
      "resdis": 1
    }
  }
}

• scope (5): Requesting a curated selection determined by the Content Owner or Marketplace.
• function (1): Using ai-all, which covers inference and generation.
• subfn (4): Specifically for agent-actions, where the AI is taking action on behalf of a user.

Content Owner Response (Package with Image & Video)

This response includes one image and one video asset, specifically detailing the sourcetype (Human vs. AI) and provenance.

{
  "package": {
    "id": "PKG-MEDIA-PREMIUM",
    "title": "Verified Travel Selection 2026",
    "seller": "premium-travel-pics.com",
    "licenseurl": "https://premium-travel-pics.com/license",
    "citation": 1,
    "scope": {
      "scope": 5,
      "ctype": [1, 2],
      "image": [
        {
          "title": ["Grand Canyon Sunset"],
          "pubdate": "2026-01-10T18:00:00Z",
          "published": 1,
          "author": ["Jane Photographer"],
          "sourcetype": 0,
          "provenance": 1,
          "provent": "c2pa.org"
        }
      ],
      "video": [
        {
          "title": ["Grand Canyon South Rim Walkthrough"],
          "dur": [45],
          "clip": 1,
          "transcript": 0,
          "sourcetype": 0,
          "provenance": 1,
          "provent": "c2pa.org"
        }
      ]
    },
    "retrieval": {
      "auth": 1,
      "endpoint": "https://media-api.premium-travel.com/v1/download",
      "type": [2]
    }
  }
}

Key Object Metadata

• image.sourcetype (0): Indicates the image was created by a human.
• video.clip (1): Denotes this is a clip derived from a longer video.
• video.dur (45): The duration of the video asset is 45 seconds.
• provenance (1): Confirms that provenance information is available for these assets.
• provent: Identifies C2PA as the entity providing the provenance.
• retrieval.auth (1): Access requires an api_key.

The AI system is asking to crawl an application with a lot of images in response to a user query that would like to create an image in the style of a well known artist. The AI System uses its issued token to request images related to a specific artist's style for generative use.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "creative-gen.ai",
    "ua": "CreativeGenBot/2.0",
    "id": "TL-REG-554433",
    "aisysuse": {
      "lid": "TOKEN-STYLE-99-ALPHA",
      "aiauth": 2,
      "uri": [
        "https://digital-archive-museum.org/collections/artist-style-ref"
      ],
      "scope": 4,
      "function": [3],
      "subfn": [5],
      "resdis": 1
    }
  }
}

• aiauth (2): Authenticating using a Token.
• scope (4): Targeting by topic (the specific artist's name or movement).
• function (3): Identified as ai-input, which includes use for generative AI responses.
• subfn (5): Categorized as other for style-transfer.
• resdis (1): Indicates the generated image will be displayed to a human.

Content Owner Response: Curated Style Reference Package

This response acknowledges the AI system's token-based request and delivers a curated selection of images—some with verified C2PA provenance and some without—along with instructions for retrieval.

{
  "package": {
    "id": "PKG-ARTIST-COLLECTION-02",
    "title": "Impressionist Style Reference Set - Mixed Provenance",
    "seller": "digital-archive-museum.org",
    "licenseurl": "https://digital-archive-museum.org/licensing/style-transfer",
    "citation": 1,
    "scope": {
      "scope": 5,
      "max": 1,
      "ctype": [2],
      "image": [
        {
          "title": ["Starry Night Interpretation"],
          "pubdate": "2026-01-05T12:00:00Z",
          "published": 1,
          "author": ["Original Artist Name"],
          "sourcetype": 0,
          "provenance": 1,
          "provent": "c2pa.org"
        },
        {
          "title": ["Abstract Study in Blue"],
          "pubdate": "2026-01-10T14:20:00Z",
          "published": 1,
          "author": ["Unknown Artist"],
          "sourcetype": 0,
          "provenance": 0
        }
      ]
    },
    "retrieval": {
      "auth": 2,
      "endpoint": "https://api.digital-archive-museum.org/v1/assets",
      "type": [2]
    }
  }
}

Metadata Breakdown

• package.citation (1): Since the AI system indicated it would display results to a human user (resdis: 1), the content owner here explicitly requires citation.
• scope.scope (5): The package is defined as a curated selection.
• scope.max (1): Indicates an upper limit on the crawl; the AI system should reference the license for the specific cap.
• Image 1 (Provenance Available): Marked as human-created (sourcetype: 0) with provenance: 1 and a provent (Provenance Entity) of c2pa.org.
• Image 2 (No Provenance): Also human-created, but marked with provenance: 0. Per the spec, the provent field is omitted because it is only required when provenance is 1.
• retrieval.auth (2): Specifies that OAuth2 is required to access the content once a token has been issued.
• retrieval.type (2): Informs the AI system that the content is accessible via an API.

The AI System would like to crawl an entire news podcast feed to train and ground its model.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "audio-intel.ai",
    "ua": "AudioIntelBot/3.0",
    "id": "TL-REG-223344",
    "aisysuse": {
      "lid": "LIC-POD-PRO-2026",
      "aiauth": 2,
      "uri": [
        "https://news-podcasts.com/daily-brief"
      ],
      "scope": 0,
      "function": [2, 3],
      "subfn": [0, 2],
      "resdis": 0
    }
  }
}

• scope (0): Requests the full corpus of the feed.
• function (2, 3): Declares use for ai-train and ai-input.
• subfn (0, 2): Specifically targets training and grounding.
• resdis (0): Indicates results will not be displayed to a human user.

Content Owner Response (Package)

The Content Owner provides the package details, including a specific Audio object that highlights the availability of a transcript—crucial for grounding tasks.

{
  "package": {
    "id": "PKG-DAILY-BRIEF-FULL",
    "title": "Daily Brief Podcast Archive",
    "seller": "news-podcasts.com",
    "licenseurl": "https://news-podcasts.com/licensing",
    "citation": 1,
    "scope": {
      "scope": 0,
      "max": 0,
      "ctype": [3],
      "audio": [
        {
          "title": ["Global Markets Update - Feb 27"],
          "dur": [1800],
          "wordcount": [4500],
          "transcript": 1,
          "pubdate": "2026-02-27T06:00:00Z",
          "published": 1,
          "author": ["Host Name", "Guest Expert"],
          "sourcetype": 0,
          "provenance": 0
        }
      ]
    },
    "retrieval": {
      "auth": 2,
      "endpoint": "https://api.news-podcasts.com/v1/rss",
      "type": [1]
    }
  }
}

Key Metadata Elements

• audio.dur (1800): Indicates the duration is 1800 seconds (30 minutes).
• audio.transcript (1): Per the specification table for Audio, 1 = Yes, a transcription is available.
• audio.wordcount (4500): Provides the count of words in the audio transcript.
• audio.sourcetype (0): Confirms the audio is human-created.
• retrieval.type (1): The content is delivered via an RSS feed.
• retrieval.auth (2): Requires OAuth2 for access.

The AI System would like to crawl just episodes of the news podcast feed published after January 1, 2026 to answer a user query about a specific news headline. The AI system requests content published after the specified date for RAG (Retrieval-Augmented Generation).

AI System Request (AISystem)

{
  "aisystem": {
    "name": "current-events-agent.ai",
    "ua": "CurrentEventsBot/1.5",
    "id": "TL-REG-112233",
    "aisysuse": {
      "lid": "LIC-RAG-2026-X",
      "aiauth": 2,
      "uri": [
        "https://news-podcasts.com/daily-brief"
      ],
      "scope": 2,
      "function": [3],
      "subfn": [1],
      "resdis": 1
    }
  }
}

• scope (2): Specific to a date range filter (e.g., content published after YYYYMMDD).
• function (3): Identified as ai-input, which includes RAG and grounding.
• subfn (1): Specifically defines the use case as RAG.
• resdis (1): Indicates that results will be displayed to a human user.

Content Owner Response (Package)

The Content Owner responds with a package containing assets published within that window. For RAG purposes, providing the transcript (1) is essential.

{
  "package": {
    "id": "PKG-POD-2026-Q1",
    "title": "Daily Brief - 2026 Post-Jan Archive",
    "seller": "news-podcasts.com",
    "licenseurl": "https://news-podcasts.com/licensing/rag-terms",
    "citation": 1,
    "scope": {
      "scope": 2,
      "max": 0,
      "ctype": [3],
      "audio": [
        {
          "title": ["Economic Outlook: February 2026"],
          "dur": [1500],
          "wordcount": [3800],
          "transcript": 1,
          "pubdate": "2026-02-15T08:00:00Z",
          "published": 1,
          "author": ["Casey Jordan"],
          "sourcetype": 0,
          "provenance": 0
        }
      ]
    },
    "retrieval": {
      "auth": 2,
      "endpoint": "https://api.news-podcasts.com/v1/feed",
      "type": [2]
    }
  }
}

Key Object Details

• package.citation (1): Required, as the AI system intends to display the result to a human.
• scope.scope (2): Acknowledges the date range scope.
• audio.transcript (1): Per the Audio object specification, 1 = Yes, a transcript is available.
• audio.pubdate: "2026-02-15T08:00:00Z" confirms the asset meets the "after January 1st" requirement.
• audio.provenance (0): Indicates no provenance is available for this specific audio asset.

The AI System is responding to a user query to create a video in Steampunk style. The AI system requests content filtered by topic (scope 4) to support a generative user query.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "creative-gen.ai",
    "ua": "CreativeGenBot/2.0",
    "id": "TL-REG-554433",
    "aisysuse": {
      "lid": "LIC-TOPIC-SP-2026",
      "aiauth": 2,
      "uri": [
        "https://video-archive.com/steampunk-search"
      ],
      "scope": 4,
      "function": [3],
      "subfn": [1, 2],
      "resdis": 1
    }
  }
}

• scope (4): Filters the request specifically to the "steampunk" topic.
• function (3): Identifies the use as ai-input for generative responses.
• subfn (1, 2): Specifies the sub-functions of RAG and grounding.
• resdis (1): Indicates the final output will be seen by a human.

Content Owner Response (Package)

The Content Owner provides a curated collection of videos with varying provenance and creation sources.

{
  "package": {
    "id": "PKG-STEAMPUNK-001",
    "title": "Steampunk Visual Archives",
    "seller": "video-archive.com",
    "licenseurl": "https://video-archive.com/legal/ai-terms",
    "citation": 1,
    "scope": {
      "scope": 4,
      "max": 1,
      "ctype": [1],
      "video": [
        {
          "title": ["History of Steampunk Aesthetics"],
          "dur": [600],
          "clip": 0,
          "transcript": 1,
          "pubdate": "2026-01-20T10:00:00Z",
          "author": ["Gearheart Media"],
          "sourcetype": 0,
          "provenance": 1,
          "provent": "c2pa.org"
        },
        {
          "title": ["AI-Generated Steampunk Landscape"],
          "dur": [60],
          "clip": 1,
          "transcript": 0,
          "pubdate": "2026-02-15T14:30:00Z",
          "author": ["StyleBot-V1"],
          "sourcetype": 1,
          "provenance": 0
        }
      ]
    },
    "retrieval": {
      "auth": 2,
      "endpoint": "https://api.video-archive.com/v1/media",
      "type": [2]
    }
  }
}

Metadata Highlights

• citation (1): Requires the AI System to cite the Content Owner.
• Video 1 (Human + Provenance):
  • sourcetype (0): Indicates the video was created by a human.
  • provenance (1): Confirms that technical provenance is available via c2pa.org.
  • transcript (1): Confirms a transcription is available for this asset.
• Video 2 (AI + No Provenance):
  • sourcetype (1): Explicitly identifies the asset as AI-generated.
  • provenance (0): Indicates no technical provenance is available for this specific file.
  • clip (1): Denotes that this asset is a clip rather than full-length content.
• retrieval.type (2): Informs the AI system the content is delivered via API.

This scenario explores the Agent-View and Agent-Actions sub-functions. This is designed for AI "Agents" that don't just consume content for knowledge, but crawl data to perform tasks like booking travel or making reservations. The AI System, acting as a personal travel agent, requests access to a restaurant's reservation site to check for availability and perform a booking on behalf of a user.

AI System Request (AISystem)

{
  "aisystem": {
    "name": "concierge-bot.ai",
    "ua": "ConciergeBot/1.0",
    "id": "TL-REG-889900",
    "aisysuse": {
      "lid": "LIC-AGENT-ACT-2026",
      "aiauth": 2,
      "uri": [
        "https://bistro-reserve.com/api/availability"
      ],
      "scope": 5,
      "function": [1],
      "subfn": [3, 4],
      "resdis": 1
    }
  }
}

• scope (5): Requesting a curated selection (the specific reservation data).
• function (1): Using ai-all, which includes model evaluation and retrieval.
• subfn (3, 4): Combining agent-view (gathering data for future action) and agent-actions (taking the action).
• resdis (1): The outcome of the action will be shown to a human user.

Content Owner Response (Package)

The Marketplace or Content Owner provides the retrieval endpoint for the Agent, typically utilizing the Model Context Protocol (MCP) or an API for structured interaction.

{
  "package": {
    "id": "PKG-RESERVE-TOOL-01",
    "title": "BistroReserve Agent Interface",
    "seller": "bistro-reserve.com",
    "licenseurl": "https://bistro-reserve.com/legal/agent-terms",
    "citation": 0,
    "scope": {
      "scope": 5,
      "max": 1,
      "ctype": [0],
      "text": [
        {
          "title": "Real-time Table Availability",
          "published": 1,
          "sourcetype": 0,
          "provenance": 0
        }
      ]
    },
    "retrieval": {
      "auth": 2,
      "endpoint": "https://mcp.bistro-reserve.com/v1/connect",
      "type": [3]
    }
  }
}

Key Metadata Elements

• scope.scope (5): Confirms the response is for a curated selection of reservation data.
• text.sourcetype (0): Confirms the data provided (availability, descriptions) is human-created.
• retrieval.type (3): Specifies MCP (Model Context Protocol), where the AI System client connects to specialized connectors to expose functionality.
• retrieval.auth (2): Requires OAuth2 for the Agent to securely interact with the booking system.
• package.citation (0): Citation is not required for this functional data exchange.