Full Specification

The AI Catalog specification is published as a W3C-style normative document.

View the full specification

What's in the specification

The specification covers:

• Core schema — formal definitions for AI Catalog, Catalog Entry, Host Info, Publisher, and all fields
• Trust Manifest — identity binding, attestations, provenance links, signing, and verification procedures
• Discovery — well-known URI, link relation headers, and agent-driven discovery
• Conformance levels — normative requirements for Minimal, Discoverable, and Trusted catalogs
• Version handling — compatibility rules for producers and consumers
• Security considerations — trust layers, catalog poisoning, typosquatting, embedded content safety
• CDDL schema — formal machine-readable schema
• Mappings — how AI Catalog maps to OCI Distribution, MCP servers (via MCP Server Cards), and Claude Code Plugins

Relationship to these docs

The developer documentation you're reading now is a practical guide aimed at getting you productive quickly. It draws from the specification but is not normative.

When building a production implementation, refer to the specification for precise requirements — especially for Trust Manifest verification, signature algorithms, and conformance obligations.