Agent Registry uses standardized terminology to describe the components, resources, and governance mechanisms across your AI ecosystem. Understanding these foundational concepts helps you register, discover, connect, and govern agents, MCP servers, endpoints, and skills in Agent Registry.
A2A skill
Compare with skill.
A capability description listed inline within an agent's agent-card.json
Agent Card. An A2A skill
describes the tasks an agent can perform and its communication interfaces so
that other agents know when and how to delegate work to it.
Agent Registry supports versions 0.3 and 1.0 of the
Agent2Agent (A2A) specification. For a comparison
with standalone skills, see
Differences between standalone and A2A skills.
Agent
An autonomous actor in your AI ecosystem that plans, reasons, invokes tools,
and collaborates with other agents. An agent is defined by its
agent identifier
and each A2A skill or standalone skill it possesses. In
Agent Registry, a registered agent is represented as a read-only Agent
discovery resource.
Agent principal
The unique Identity and Access Management (IAM) identity assigned to an agent, letting it hold permissions, access downstream resources, and generate audit logs. It takes the form of a verifiable Google Cloud service account or managed workload identity, such as a SPIFFE ID, bound directly to the compute runtime of the agent.
Principal strings incorporate the runtime reference path and support three scopes: the engine instance, project-wide principal sets, and organization-wide principal sets.
For complete syntax and access control guidance, see Agent principals.
Agentic component
A modular, functional entity within an AI ecosystem that Agent Registry lets you register, discover, and govern. The registry categorizes agentic components into four primary types: agent, MCP server, endpoint, and skill.
For more information, see Component lifecycle and discovery.
Agentic workflow
A sequence of steps where an AI agent independently determines tool usage, reasoning, and execution paths to accomplish a task.
Auth provider
A configuration within Agent Identity auth manager to store, acquire, and manage credentials, such as API keys or OAuth tokens, for connecting your agent to external tools and downstream services.
For more information about binding agents to auth providers, see Component relationships.
Binding
A directed connection between a source agent and a target resource, such as another agent, an MCP server, or an endpoint. Bindings establish explicit delegation pathways between components and associate an agent with an auth provider for delegated permissions.
For more information, see Bindings.
Data resource
A specific data context or dataset exposed by an MCP server that an agent can access to ground its responses or inform its actions.
Discovery
The process of querying Agent Registry to find a registered agent, MCP tool, endpoint, or skill using natural language search, tags, or registry identifiers.
Discovery in Agent Registry focuses on consumption discovery, which identifies the capabilities and components that your orchestrator agents can invoke at runtime. For more information, see Discovery types.
Endpoint
A target URL, typically a REST API, accessed by an agent. In
Agent Registry, a registered endpoint is represented as a read-only Endpoint
discovery resource, letting you centrally govern which external services an
agent can access.
Governance project
The Google Cloud project where you enable the Agent Registry API to centrally catalog, discover, and govern your agentic components. For more information about single-project and cross-project configurations, see Define the governance project for a central registry.
Governance scope
The geographic and architectural boundary where Agent Registry resources are registered and governed.
Agent Registry uses a central Google Cloud governance project, where you enable the Agent Registry API to catalog and govern your agentic components. Automatic registration discovers supported workloads deployed in that governance project. To govern remote components, which are agents, MCP servers, or endpoints hosted outside the governance project in separate Google Cloud projects, on-premises environments, or external clouds, you manually register them in the governance project's registry.
When Agent Registry automatically registers a Google or Google Cloud
remote MCP server, it creates the McpServer resource only in the global
location, not in individual regions. You can manually register agents, MCP
servers, and endpoints, and create bindings, in a supported regional location
or the global location. Regional gateways can govern and route traffic to
both regional resources and global MCP servers.
For full regional scoping rules and policy requirements, see Agent Registry locations.
Model Context Protocol (MCP)
The open standard used to connect AI models to data sources and tools, replacing bespoke plugins.
MCP server
A program that implements the
Model Context Protocol (MCP)
to expose standardized tool and data resource
capabilities to an AI agent. In Agent Registry, a registered MCP
server is represented as a read-only McpServer discovery resource.
Publisher
An entity that creates and registers a skill, represented as a
Publisher resource in Agent Registry. A publisher establishes the ownership
and
skill identifier
syntax for a skill.
Google-managed skills use service domains. Manually created skills are
automatically scoped to the private publisher with a private- prefix.
For more information, see Publishers.
Registration
The process of adding an agentic component, such as an agent, MCP server, endpoint, or skill, to Agent Registry.
Agent Registry supports two registration models:
- Automatic registration: automatic ingestion of agents, MCP servers, and Google-managed skills from supported Google Cloud runtimes, such as Agent Runtime on Gemini Enterprise Agent Platform, deployed in the governance project where Agent Registry is enabled.
- Manual registration: explicit registration of remote or custom agents, MCP servers, endpoints, and private standalone skills, including components hosted in separate Google Cloud projects, on-premises environments, external clouds, or unsupported runtimes, by creating resources in the governance project's registry.
For a detailed architectural comparison, see Registration models.
Registry identifier
A globally unique, immutable Uniform Resource Name (URN) that Agent Registry assigns to a registered agentic component. This identifier provides a stable reference for consumers and orchestrator agents to discover, filter, and search components in Agent Registry, remaining constant regardless of underlying infrastructure changes or code updates.
For detailed URN syntax formats, see Registry identifiers.
Runtime reference
The physical runtime location or compute infrastructure hosting an
agent, MCP server, or endpoint, represented
in Agent Registry by the
agentregistry.googleapis.com/system/RuntimeReference attribute. For example,
a runtime reference can point to an
Agent Runtime reasoning engine, a
Google Kubernetes Engine (GKE) deployment, or a Cloud Run service.
The runtime reference path is embedded within the agent principal for IAM policy evaluation. For more information, see Runtime references.
Service resource
The writable resource in Agent Registry, identified by its own
resource name, that
you use for management operations when you manually register an
agent, an MCP server, or an endpoint. The
Service resource stores the component's metadata, including the type of
agentic component and configuration details such as its interface URL,
description, protocol binding, and capability specification. An automatically
registered component or standalone skill doesn't use a Service
resource.
You create, update, or delete a Service resource to register, update metadata
for, or unregister a manually registered component in Agent Registry without
modifying or deleting the underlying deployment. When you create or update a
Service resource, Agent Registry reads the metadata and automatically
creates or updates the corresponding Agent, McpServer, or Endpoint
discovery resource. In Agent Registry, consumers, such as orchestrator agents,
client applications, and developers, use these read-only discovery resources to
list, search, and view agentic components. For more information, see
Agentic component representation in Agent Registry.
Skill
Compare with A2A skill.
A standalone package registered as a top-level Skill resource in
Agent Registry that extends the capabilities of an agent by
providing specific instructions and optional scripts to complete a task. A skill
contains a SKILL.md instruction file, optional scripts, and metadata
assets.
A skill is identified by a skill identifier scoped by a publisher and a standard resource name. Each skill maintains a default revision pointer and has one or more versioned snapshots, each represented as a skill revision.
For architecture, packaging, and lifecycle details, see Skills architecture.
Skill revision
An immutable, versioned snapshot of a skill package, represented as a
SkillRevision resource in Agent Registry. A skill revision lets you update
skill instructions or optional scripts without changing the parent
skill identifier
or breaking existing agent dependencies.
For more information about revision versioning and immutability, see Skill revisions.
Tool
A deterministic function provided by an MCP server that an agent can invoke.