Supercharge your AI brokers: The New ADK Integrations Ecosystem

An AI agent is just as succesful because the methods it could possibly work together with. An agent that may ‘suppose’ is nice, however an agent that may handle code repositories, set off workflows, or question databases is transformative.

This shift from chat to motion is precisely what Agent Growth Equipment (ADK) was constructed for. As an open-source framework, ADK offers the versatile primitives builders have to construct, check, and deploy these sorts of advanced, production-grade agentic workflows.

That’s why we’re thrilled to announce a major enlargement to the Agent Growth Equipment (ADK) ecosystem. We now have partnered with main developer platforms to convey a broad suite of third-party instruments and integrations immediately into ADK.

With only a few strains of code, these integrations enable your brokers to work together seamlessly with the actual world. For instance, watch how simply an agent can search Hugging Face to find fashions and generate code:

Listed below are the associate integrations obtainable immediately within the ADK Instruments and Integrations documentation:

Code & growth

Brokers can join immediately into your growth instruments that can assist you handle code, check APIs, and execute scripts in safe environments:

Daytona: Execute code, run shell instructions, and handle recordsdata in remoted sandboxes.
GitHub: Analyze code, handle points and PRs, and automate workflows.
GitLab: Carry out semantic code search, examine CI/CD pipelines, and handle merge requests.
Postman: Handle API collections, check APIs, and generate shopper code.
Restate: Resilient agent execution with sturdy classes and automated restoration.

Mission administration

Give your brokers entry to your crew’s venture monitoring and documentation:

Asana: Handle tasks, duties, and targets for crew collaboration.
Atlassian: Handle Jira points, search Confluence pages, and replace crew content material.
Linear: Handle points, monitor venture cycles, and streamline growth workflows.
Notion: Search workspaces, create pages, and handle duties and databases.

Databases

Join your brokers to databases and vector search engines like google and yahoo:

Chroma: Retailer paperwork and retrieve data utilizing semantic search and metadata filtering.
MongoDB: Question collections, handle databases, and analyze schemas.
Pinecone: Retailer information, carry out semantic search, and rerank outcomes throughout vector indexes.

Reminiscence

Give your brokers persistent, searchable reminiscence throughout conversations:

GoodMem: Computerized or agent-controlled semantic reminiscence with multimodal doc assist.
Qdrant: Semantic reminiscence layer for storing and retrieving data with vector search.

Observability

Monitor, debug, and optimize your brokers in manufacturing:

AgentOps: Session replays, metrics, and monitoring for ADK brokers.
Arize AX: Manufacturing-grade observability, debugging, and analysis of LLM functions.
Freeplay: Finish-to-end observability, immediate administration, and analysis for AI brokers.
MLflow: Ingest OpenTelemetry traces for agent runs, instrument calls, and mannequin requests.
Monocle: Open-source observability, tracing, and debugging of LLM functions.
Phoenix: Open-source, self-hosted observability, tracing, and analysis of LLM functions.
W&B Weave: Log, visualize, and analyze mannequin calls and agent efficiency.

Connectors

Join your brokers to exterior apps and providers by means of integration platforms:

n8n: Set off automated workflows, join apps, and course of information.
StackOne: Join brokers to 200+ SaaS suppliers by means of a unified integration gateway.

AI fashions & datasets

Give your brokers direct entry to the AI/ML ecosystem:

Hugging Face: Entry fashions, datasets, analysis papers, and 1000’s of Gradio AI functions.

Funds

Let your brokers work with you to deal with your monetary operations and billing:

PayPal: Handle funds, ship invoices, and deal with subscriptions.
Stripe: Handle funds, prospects, subscriptions, and invoices.

Speech & audio

Add voice and audio capabilities to your brokers:

Cartesia: Generate speech, localize voices throughout languages, and create audio content material.
ElevenLabs: Generate speech, clone voices, transcribe audio, and create sound results.

E mail and messaging

Let your brokers ship, obtain, and handle electronic mail:

AgentMail: Devoted electronic mail inboxes for brokers to handle conversations and attachments.
Mailgun: Ship emails, monitor supply metrics, and handle mailing lists.

ADK additionally contains built-in integrations with Google Cloud providers together with BigQuery, Spanner, Pub/Sub, and extra. See the total listing within the ADK Instruments and Integrations documentation.

A unified developer expertise

Your agent’s core logic should not be coupled to its instruments. Through the use of the McpToolset primitive or plugin structure, you may add any of those third-party integrations into your agent with only a few strains of code. ADK lets you experiment and discover one of the best instrument for the job with out ever having to refactor your agent. You needn’t fear in regards to the underlying implementation particulars – simply copy the configuration for the instrument you need, and the ADK handles the remaining.

Right here is an instance of including the GitHub integration to an ADK agent in Python, which immediately helps you empower your agent to handle repositories and automate workflows:

from google.adk.brokers import Agent
from google.adk.instruments.mcp_tool import McpToolset
from google.adk.instruments.mcp_tool.mcp_session_manager import StreamableHTTPServerParams

GITHUB_TOKEN = "YOUR_GITHUB_TOKEN"

# Initialize your agent
root_agent = Agent(
    mannequin="gemini-3-flash-preview",
    identify="github_agent",
    instruction="Assist customers get data from GitHub",
    # Add a third-party integration by configuring McpToolset
    instruments=[
        McpToolset(
            connection_params=StreamableHTTPServerParams(
                url="https://api.githubcopilot.com/mcp/", 
                headers={
                    "Authorization": f"Bearer {GITHUB_TOKEN}",
                    "X-MCP-Toolsets": "all",
                    "X-MCP-Readonly": "true"
                },
            ),
        )
    ],
)

Python

And right here’s what the ADK agent appears to be like like in motion because it queries and searches by means of GitHub: