An AI agent that can reason is useful. An AI agent that can open a pull request, update a Jira ticket, query a database, and trigger a workflow is something else entirely. That’s the shift Google is making with a major expansion to its Agent Development Kit (ADK), announced February 27.
The update brings third-party integrations directly into the open-source framework, connecting ADK agents to development tools, project management platforms, databases, observability systems, and workflow automation services. The intent is clear: Move AI agents from conversation partners to active participants in engineering workflows.
What’s New
ADK has been open-source since Google Cloud NEXT 2025. It supports Python, TypeScript, Java, and Go. While optimized for Gemini, it’s model-agnostic through LiteLLM integration with 100+ providers, including Anthropic and OpenAI.
The new integrations add direct connections across five categories that map to how DevOps teams actually work.
Development and code tools. GitHub integration covers pull requests, issues, and workflows. GitLab handles CI/CD inspection and merge management. Postman enables API test automation. Daytona provides sandboxed code execution. And Restate supports durable, resumable agent sessions — important for long-running tasks that can’t afford to lose state.
Project and work management. Asana, Atlassian (Jira and Confluence), Linear, and Notion integrations let agents track issues, update tickets, and search or create documentation. An agent can participate in daily team workflows rather than just summarize them.
Data and memory. Chroma, MongoDB, and Pinecone enable database queries and vector semantic search. GoodMem and Qdrant add persistent memory so agents maintain context across sessions — critical for agents handling ongoing work rather than one-off tasks.
Observability. This category matters most for production deployments. AgentOps provides session replays and metrics. Arize AX handles production-grade debugging. MLflow ingests OpenTelemetry traces for agent runs and tool calls. Phoenix and Monocle offer open-source tracing. Weights & Biases Weave logs model calls and agent performance. Freeplay adds end-to-end prompt management and evaluation.
Workflow automation. n8n connects agents to automated workflows across apps. StackOne provides a unified gateway to 200+ SaaS providers.
AI/ML ecosystem. Hugging Face integration gives agents access to models, datasets, research papers, and Gradio applications.
The integration model uses ADK’s McpToolset primitive and plugin architecture. Adding any of these tools requires a few lines of configuration code. The agent’s core logic remains decoupled from its tools, so teams can swap integrations without refactoring the agent’s logic.
“Agent accountability can only be achieved through observability-native across development workflows and the agent control plane. Google’s ADK expansion moves AI agent frameworks inside these engineering workflows. Seven observability integrations at launch, including OpenTelemetry support through MLflow, reflect vendors competing to own the agent control-plane layer, where telemetry and governance are embedded from design rather than bolted on later. The curated ecosystem model trades MCP’s flexibility for reliability, a distinction that will shape enterprise confidence for production deployments,” per Mitch Ashley, VP and practice lead for software lifecycle engineering at The Futurum Group.
“For DevOps teams, the execution obligation is stack alignment. ADK provides a direct path for your toolchain to run GitHub, Jira, and MongoDB. Where it doesn’t, MCP-based integrations remain the more flexible option. Either way, observability requirements must be established before agents touch production workflows, because agent behavior is non-deterministic, and you cannot govern what you cannot see.”
Why This Matters for DevOps
The observability category stands out. Seven providers at launch — including OpenTelemetry support through MLflow — suggest Google recognizes that monitoring AI agents in production is fundamentally different from monitoring traditional applications. Agent behavior is non-deterministic. The same input can produce different tool call sequences.
Having this built into the framework from day one sets ADK apart from frameworks where observability is an afterthought.
The development tool integrations also address a practical gap. Right now, most AI coding agents — Claude Code, Cursor, GitHub Copilot — operate as standalone tools. ADK’s GitHub and GitLab integrations let agents participate in repository workflows programmatically: opening PRs, managing issues, and inspecting CI/CD pipelines. Combined with Daytona’s sandboxed execution and Restate’s durable sessions, you get the building blocks for agents that handle end-to-end multi-step development workflows.
The project management integrations add another dimension. An agent that can read a Jira ticket, check the relevant repository, run tests, and update the ticket with results connects the planning layer to the execution layer without a human having to shuttle context between tools.
The Competitive Context
Google isn’t alone here. Anthropic’s Model Context Protocol (MCP) has become a de facto standard for connecting AI agents to external tools, and most major frameworks — including ADK — support it. Cursor, Claude Code, and OpenClaw all use MCP.
What ADK brings is a curated, tested ecosystem with official partnerships rather than a bring-your-own-integration model. The tradeoff is familiar: a managed ecosystem offers reliability at the cost of flexibility. An open protocol offers flexibility at the cost of variable integration quality.
For DevOps teams, the question is whether the specific tools in the ADK ecosystem align with their stack. If you’re running GitHub or GitLab, Jira or Linear, MongoDB or Pinecone — the path is straightforward. If your toolchain lives elsewhere, MCP-based integrations remain the more flexible option.
What to Watch
The ADK expansion signals where Google thinks AI agent development is heading: Away from chatbots and toward systems that operate inside existing engineering infrastructure. The framework now covers code management, project tracking, data access, observability, and workflow automation — with deployment flexibility to run locally, in containers, or on Google Cloud.
Documentation and code examples are available at Google’s ADK Tools and Integrations page. ADK supports Gemini 3 Pro, Gemini 3 Flash, and 100+ third-party models through LiteLLM.