Google Cloud Agent Registry Integration¶

The AgentRegistry client in the ADK provides a high-level interface for interacting with the Google Cloud Agent Registry service. This allows you to discover and integrate with registered agents and other resources within your Google Cloud project.

Initializing the AgentRegistry Client¶

To get started, you need to initialize the AgentRegistry client with your Google Cloud project ID and location.

from google.adk.integrations.agent_registry import AgentRegistry

agent_registry = AgentRegistry(
    project_id="your-gcp-project-id",
    location="your-gcp-location",
)

Working with Endpoints¶

The AgentRegistry client now includes methods for discovering and resolving model endpoints.

Listing Endpoints¶

You can list all available endpoints in your configured project and location using the list_endpoints() method.

all_endpoints = agent_registry.list_endpoints()
print(all_endpoints)

Getting a Specific Endpoint¶

If you know the name of a specific endpoint, you can retrieve its details with the get_endpoint() method.

endpoint_details = agent_registry.get_endpoint("projects/your-gcp-project-id/locations/your-gcp-location/endpoints/your-endpoint-id")
print(endpoint_details)

Resolving Model Names¶

The get_model_name() method simplifies the process of extracting a model resource name from an endpoint.

model_name = agent_registry.get_model_name("projects/your-gcp-project-id/locations/your-gcp-location/endpoints/your-endpoint-id")
print(model_name)

Authenticating MCP Toolsets¶

When retrieving an MCP toolset with get_mcp_toolset(), you can now provide authentication details using the auth_scheme and auth_credential parameters. This allows the ADK to handle authentication when communicating with the MCP server.

from google.adk.auth import AuthScheme
from google.adk.auth.auth_credential import ServiceAccountCredential

# Example using a service account for authentication
auth_credential = ServiceAccountCredential(
    service_account_email="your-service-account@your-project.iam.gserviceaccount.com"
)

mcp_toolset = agent_registry.get_mcp_toolset(
    mcp_server_name="projects/your-gcp-project-id/locations/your-gcp-location/mcpServers/your-mcp-server",
    auth_scheme=AuthScheme.GOOGLE_ID_TOKEN,
    auth_credential=auth_credential
)

Custom HTTPX Client for Remote A2A Agents¶

For advanced use cases, such as adding custom logging or telemetry, get_remote_a2a_agent() now supports injecting a custom httpx.AsyncClient.

import httpx

# Create a custom httpx client
custom_client = httpx.AsyncClient(timeout=30.0)

remote_agent = agent_registry.get_remote_a2a_agent(
    agent_name="projects/your-gcp-project-id/locations/your-gcp-location/agents/your-agent-id",
    httpx_client=custom_client
)

This provides greater flexibility for controlling the HTTP communication between the ADK and the remote agent.