DCAT-AP-UK: DCAT Application Profile for the UK
Help Site map

Adopt

AI-ready metadata

Updated 3 sections on this page are new or updated since 1.2.0-rc.1 (2026-11-06). Everything that changed

AI-ready metadata answers what an AI team, or an AI system, needs before using data: what it is, whether it may be used for training, how good it is, what it covers, and under what policy. It also lets AI assistants read the profile itself.

What AI-ready means here

Described for machines

Every record converts to JSON-LD, Turtle, schema.org and Croissant. The profile publishes its own shapes, schema and context.

Permission to train, stated

The AI-ready extension makes the owner say whether the data may train, fine-tune or evaluate models. A usage policy says it for machines.

Quality you can see

A quality score on five dimensions, and a link to the data quality report, so models are not trained on data nobody can judge.

An assistant that knows the profile

An MCP server lets AI assistants look up properties, check records and convert them, using the same rules as the shapes.

A map for language models

llms.txt and llms-full.txt at the site root point language models at the profile's content in plain markdown.

Honest limits

The use for AI is the owner's statement. The licence and the law still bind. An AI system must not treat a missing statement as permission.

It follows the Guidelines and best practices for making government datasets ready for AIV.

The AI-ready extension

Makes the record answer the questions an AI team asks first: may we train on it, is it synthetic, where is the data card, what policy applies, how good is it, and what period does it cover. See the extension.

What the AI-ready extension adds
PropertyClassObligationTermChangeDefinition
Use for AIDatasetMandatory Provisionaldcatuk:aiUseNewWhether the data may be used to train, fine-tune or evaluate AI models.
Personal dataDatasetMandatorydcatuk:personalDataRaised from recommendedWhether the data contains information about identifiable living people, and in what form.
Usage policyDatasetRecommendedodrl:hasPolicyRaised from optionalA machine-readable policy, in the W3C Open Digital Rights Language (ODRL), saying what may and may not be done with the resource.
Data cardDatasetRecommended Provisionaldcatuk:dataCardNewA page that documents how the data was collected, what it contains, what it is for and its known gaps and biases.
Synthetic dataDatasetRecommended Provisionaldcatuk:syntheticDataNewWhether the data was generated rather than collected.
Data qualityDatasetRecommendeddqv:hasQualityAnnotationRaised from optionalA link to a published statement or measurement of the quality of the data.
Time period coveredDatasetRecommendeddct:temporalNarrowedThe period of time the data is about.
DocumentationDatasetRecommendedfoaf:pageRaised from optionalA web page that documents the resource, such as a user guide or a methodology.
Generated byDatasetOptionalprov:wasGeneratedByNewThe process, model or system that generated the data, as an identifier.

Usage policies machines can readUpdated

Link a policy with usage policy. Each template is ODRL 2.2 in JSON-LD. Purposes such as training AI models are terms this profile adds, and are provisional. Where a licence and a policy disagree, the licence wins. ODRL Information Model 2.2V

Usage policy templates (ODRL 2.2)
PolicyTypeFile
Open Government Licence v3.0, as a policy: use, copy, share and adapt, with attributionSetogl-open.jsonld
Open use with attribution, but not for training AI modelsSetogl-no-ai-training.jsonld
Use only to evaluate AI models, not to train themSetai-evaluation-only.jsonld
Use for research only, without sharing furtherSetresearch-only.jsonld
An offer: use under a data sharing agreementOfferagreement-required.jsonld

Exports for search engines and machine learningUpdated

The record builder and the MCP server export any record as schema.org JSON-LD and as Croissant 1.1. The Croissant export is dataset level only: add the record sets (columns and their types) with Croissant tools. Croissant 1.1S

A map for language modelsUpdated

llms.txt lists the portal’s pages and files for language models, as the llms.txt proposal suggests; llms-full.txt holds every class, property and codelist as plain markdown. The llms.txt proposalV

An MCP server for AI assistants

A small Model Context Protocol server lets an AI assistant look up the profile and check, score, convert and export records. It runs on your machine, uses the portal’s own validator, makes no network requests and writes no files. The repository’s tests start it and call its tools.

python3 -I tools/build.py
node integrations/mcp/server.js
Tools the MCP server offers
ToolWhat it does
profile_summaryWhat DCAT-AP-UK is, its version and status, its classes and its sector extensions.
describe_classEvery property of a class with its obligation (M, R, O, C) and cardinality, optionally with extensions applied.
describe_propertyThe definition, guidance, kind, codelist and example of a property, and the classes that use it.
list_codelistThe values of a codelist, with their identifiers.
searchFind properties, classes and codelists whose name, label or definition contains the words given.
validate_recordCheck a record against DCAT-AP-UK (and any extensions). Returns errors, warnings, the conformance level and the quality score.
convert_memConvert a record in the xGov Metadata Exchange Model form to DCAT-AP-UK, listing what changed, and check it.
convert_openapiDraft a DCAT-AP-UK data service record from an OpenAPI description (parsed JSON). Email addresses are never copied.
export_recordExport a record as Turtle, schema.org JSON-LD or Croissant 1.1 (dataset level).

The server and how to add it to an MCP client

Search the portal

Keyboard shortcuts

/ or Ctrl K
Search
?
These shortcuts
Esc
Close menus, tooltips and dialogs
Arrow keys
Move between tabs and search results