a2a-protocol-consumer

a2a-protocol-consumer

Make a remote A2A-exposed agent callable from your local ADK 2.0 agent.

When to use

• A specialist agent runs in another service / language / team
• Microservices: keep agents independently deployable
• Cross-org agent sharing via published agent cards

Template — as a sub-agent

from google.adk.agents import LlmAgent
from google.adk.a2a import RemoteA2AAgent

# Reference a remote A2A agent by its agent card URL
remote_researcher = RemoteA2AAgent(
    name="remote_researcher",
    agent_card_url="https://research.example.com/.well-known/agent-card",
    auth={"type": "bearer", "token_env": "REMOTE_A2A_TOKEN"},
)

coordinator = LlmAgent(
    name="coordinator",
    model="gemini-2.5-flash",
    description="Local coordinator that delegates research to a remote A2A agent.",
    instruction="When the user asks for research, delegate to remote_researcher.",
    sub_agents=[remote_researcher],
)

root_agent = coordinator

Template — as a tool

If you'd rather call the remote agent as a tool (single-shot, no delegation):

from google.adk.a2a import a2a_tool

remote_search_tool = a2a_tool(
    name="remote_search",
    agent_card_url="https://research.example.com/.well-known/agent-card",
    description="Run a research query against the external research agent.",
)

root_agent = LlmAgent(
    name="local",
    model="gemini-2.5-flash",
    instruction="Use remote_search for any web research.",
    tools=[remote_search_tool],
)

Discovery

The agent card URL is a JSON document at /.well-known/agent-card describing the remote agent's name, description, capabilities, and auth requirements. ADK reads it on construction.

Validation

• curl https://remote/.well-known/agent-card returns valid JSON before wiring
• Auth tokens stored in env, never committed
• Network egress allowed from your runtime to the remote endpoint
• Timeouts configured (default 30s) — set explicitly for slow remotes

See also

• a2a-protocol-exposer — publish your own ADK agent over A2A