How to Connect Google ADK Agents to Dobby: 3 Integration Paths

Google's Agent Development Kit (ADK) has become one of the fastest-growing agent frameworks in 2026, with 17.9k GitHub stars and support for Python, TypeScript, Go, and Java. ADK makes it easy to build multi-agent systems with native MCP and A2A protocol support.

But building agents is only half the battle. Production agents need guardrails: cost tracking, approval gates, content safety, and a kill-switch for when things go wrong. That's where connecting ADK to a control plane like Dobby comes in.

Why ADK Agents Need a Control Plane

ADK handles the 'build' part excellently. But in production, teams face real challenges that no agent framework solves alone.

• Cost visibility: How much does each agent spend on LLM calls per day?
• Safety: What happens when an agent encounters a prompt injection attack?
• Control: Can you instantly stop an agent that's behaving unexpectedly?
• Compliance: Do you have an audit trail of every agent action?
• Approval: Should destructive actions require human review?

Dobby adds all of these capabilities without changing how your ADK agents work.

Path 1: MCP Gateway (2 Minutes)

The fastest integration. ADK has native MCP support via MCPToolset, so your agent gets access to 90 Dobby MCP tools with a few lines of code.

from google.adk import Agent
from google.adk.tools import MCPToolset

# Connect ADK agent to Dobby's MCP Gateway
dobby_tools = MCPToolset(
    command="npx",
    args=["-y", "@dobbyai/mcp-external"],
    env={
        "DOBBY_API_URL": "https://dobby-ai.com",
        "DOBBY_API_KEY": "gk_user_your_key_here",
        "DOBBY_TENANT_ID": "tenant_your_id",
    }
)

agent = Agent(
    model="gemini-2.5-flash",
    name="my-dobby-agent",
    instruction="You are a helpful agent with access to Dobby AI tools.",
    tools=[dobby_tools],
)

That's it. Every LLM call your agent makes is now tracked with cost, tokens, and latency. You'll see it in the FinOps dashboard immediately.

Path 2: A2A Protocol (Agent Collaboration)

ADK agents auto-generate Agent Cards, and Dobby supports the A2A protocol natively. This enables bidirectional collaboration: your ADK agent can delegate tasks to Dobby's agents, and Dobby can delegate tasks to yours.

• Deploy your ADK agent with a public URL
• In Dobby, go to External Agents and click Auto-Discover
• Dobby fetches the Agent Card and registers the agent
• Task delegation flows both ways with full governance

Dobby is listed on the Google Cloud AI Agent Marketplace as an A2A-compatible control plane, making it the natural complement to any ADK deployment.

Path 3: External Agent (Webhook Triggers)

For scheduled or event-driven agents. Register your ADK agent's endpoint in Dobby, set up cron schedules or manual triggers, and add approval gates before execution.

This is ideal for agents that run periodic tasks (nightly reports, weekly audits) or need human review before proceeding with destructive actions.

What You Get

• Cost tracking per agent, per provider, per model
• Content Shield (26 patterns for prompt injection detection)
• DLP (10 PII patterns: SSN, credit cards, emails, etc.)
• Kill-switch (instantly halt any agent)
• Budget limits with automatic enforcement
• Approval gates for sensitive operations
• Full audit trail (365-day retention)
• Agent Fleet dashboard with health monitoring

Getting Started

The MCP Gateway path is the fastest way to get started. Create a free Dobby account, generate a gateway key, and add the MCPToolset code above to your ADK agent. You'll see cost data in the dashboard within seconds.