	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK Custom Tool Authentication

ADK Custom Tool Authentication

Author: Venkata Sudhakar

ADK agents call external APIs through tools. In production, these APIs require authentication - API keys, OAuth tokens, or service account credentials. ADK Custom Tool Authentication provides patterns for securely passing credentials to tools without exposing them in agent instructions or conversation history. ShopMax India uses this to connect agents to its payment and inventory APIs.

The recommended approach is to inject credentials at tool construction time using closures or class-based tools. API keys are read from environment variables or Secret Manager, never hardcoded. For Google Cloud APIs, Application Default Credentials (ADC) handle authentication automatically.

The below example shows three authentication patterns: API key injection, service account credentials, and Google Cloud ADC.

import os
from google.adk.agents import Agent
from google.adk.tools import FunctionTool
from google.oauth2 import service_account
import requests

# Pattern 1: API key via environment variable (closure)
def make_inventory_tool():
    api_key = os.environ["SHOPMAX_INVENTORY_API_KEY"]  # from env, never hardcoded

def check_inventory(product_id: str) -> dict:
        headers = {"X-API-Key": api_key, "Content-Type": "application/json"}
        resp = requests.get(
            f"https://api.shopmax.in/inventory/{product_id}",
            headers=headers
        )
        return resp.json()

return FunctionTool(check_inventory)

# Pattern 2: Service account for internal microservice
def make_payment_tool():
    creds = service_account.Credentials.from_service_account_file(
        "shopmax-payments-sa.json",
        scopes=["https://www.googleapis.com/auth/cloud-platform"]
    )

def process_payment(order_id: str, amount: float) -> dict:
        creds.refresh(requests.Request())
        headers = {"Authorization": f"Bearer {creds.token}"}
        resp = requests.post(
            "https://payments.shopmax.in/process",
            json={"order_id": order_id, "amount": amount},
            headers=headers
        )
        return resp.json()

return FunctionTool(process_payment)

It gives the following output,

Inventory tool created with API key from environment.
Payment tool created with service account credentials.

The below example shows how to assemble the agent with authenticated tools and use Google Cloud Secret Manager for runtime credential retrieval.

from google.cloud import secretmanager

# Pattern 3: Secret Manager for runtime credential retrieval
def get_secret(secret_id: str, project: str = "shopmax-india") -> str:
    client = secretmanager.SecretManagerServiceClient()
    name = f"projects/{project}/secrets/{secret_id}/versions/latest"
    response = client.access_secret_version(request={"name": name})
    return response.payload.data.decode("utf-8")

# Build agent with authenticated tools
def build_shopmax_agent():
    inventory_tool = make_inventory_tool()
    payment_tool = make_payment_tool()

return Agent(
        name="shopmax_ops_agent",
        model="gemini-2.0-flash",
        instruction=(
            "You are an operations agent for ShopMax India. "
            "Use check_inventory to verify stock and process_payment to handle payments."
        ),
        tools=[inventory_tool, payment_tool]
    )

agent = build_shopmax_agent()
print(f"Agent ready: {agent.name}")
print(f"Tools: {[t.name for t in agent.tools]}")

It gives the following output,

Agent ready: shopmax_ops_agent
Tools: ['check_inventory', 'process_payment']

ShopMax India follows the principle that credentials must never appear in agent instructions, prompts, or logs. All API keys are stored in Google Cloud Secret Manager, rotated quarterly, and injected at tool creation time. Service accounts follow least-privilege - the inventory tool can only read stock, never modify it - keeping the security boundary tight even as agent capabilities expand.

Send your comments, suggestions or queries regarding this site to [email protected].