Apicurio Registry stores the contracts AI systems depend on (agent cards, MCP tool and server definitions, prompt templates and model schemas) as versioned, validated artifacts, and lets agents discover them through standard endpoints. It applies the same governance it already gives Avro, Protobuf and OpenAPI: version history, validity and compatibility rules, access control and an audit trail.
A schema registry exists because producers and consumers must agree on a contract, and that contract changes over time. Multi-agent systems have the same problem. An orchestrator needs to know which agents exist and which skills they offer; an LLM client needs the input schema of each MCP tool; a team needs to know which version of a prompt is running in production. When those contracts live in code or in ad hoc configuration, a renamed tool parameter or a removed agent skill breaks callers at runtime.
| Schema registry concept | AI equivalent in Apicurio Registry |
|---|---|
| API and event contracts | Agent cards, MCP tool and server definitions |
| Schema validation | Validation of agent cards, tool input schemas, prompt templates and model schemas |
| Compatibility rules | Breaking-change detection on MCP tool input schemas and other AI artifacts |
| Service discovery | Agent discovery over /.well-known/ endpoints and the AI Catalog |
| Governance and audit | Versioned history, labels, access control |
Store A2A agent cards as AGENT_CARD artifacts and search them by name, skill, capability and
input or output mode through /.well-known/agents. Apicurio Registry also publishes its own agent card
at /.well-known/agent.json. Read What is an agent registry?
or the A2A features documentation.
The Apicurio Registry MCP server lets an LLM client such as Claude Desktop search, read and create artifacts through
the Model Context Protocol. Separately, Apicurio Registry stores MCP tool definitions as MCP_TOOL
artifacts, and version 3.4 adds a Developer Preview of the MCP Registry API for publishing MCP server definitions.
Read What is an MCP registry? or the
MCP server integration documentation.
PROMPT_TEMPLATE artifacts hold prompt templates with typed variables, and Apicurio Registry can render
them on the server. MODEL_SCHEMA artifacts describe the inputs and outputs of AI and ML models. Both are
validated and versioned like any other artifact. See the
LLM artifact types guide.
Apicurio Registry can publish a read-only catalog of its agent cards and MCP tools at
/.well-known/ai-catalog.json, and supports search through Agentic Resource Discovery (ARD) endpoints.
See AI Catalog and ARD discovery.
apicurio.features.experimental.enabled=true plus the feature's own flag. The AI Catalog and ARD
endpoints are disabled by default. Check the documentation for your version before relying on them in production.