The era of hardcoding every AI agent into a monolithic Python script is ending. A new open-source platform called apowerb, released under the Apache-2.0 license, introduces a database-driven approach to Google’s Agent Development Kit (ADK). By storing agent configurations as database records, apowerb allows developers to create, edit, and run ADK agents via REST API calls, effectively decoupling agent logic from the deployment cycle.

From Database Rows to Live Agents

At its core, apowerb treats an agent as a simple row containing a name, instruction, model string, and tool definitions. A function named to_agent() dynamically converts these records into functional ADK objects, such as LlmAgent, SequentialAgent, or ParallelAgent. This architecture supports recursive sub-agent structures and wraps model strings in ADK’s LiteLlm, ensuring compatibility with any LiteLLM provider, including Gemini models prefixed with gemini/.

Executing Agents via HTTP and MCP Integration

The platform’s standout feature is its seamless integration with the Model Context Protocol (MCP). In a demonstration using DeepWiki’s public MCP server, developers created a 'docs_assistant' agent entirely through HTTP POST requests. The configuration specified the MCP server URL and transport type, allowing the agent to access external tools like ask_wiki_question without any local code changes. The execution flow remains standard ADK, but the entry point is a simple API call to /api/adk/run, returning a list of ADK events.

Operational Visibility and Team Scalability

While ADK handles the agent loop, apowerb adds the operational layer teams desperately need. The platform provides a web UI and REST API for agent management, eliminating the need for redeployments when instructions change. It also introduces critical enterprise features such as per-user token quotas, per-session audit trails, and support for scheduled or email-triggered runs. Self-hosting is streamlined via Docker Compose or Helm charts, making it viable for production environments.

Key Takeaways

  • apowerb enables dynamic agent creation and management via REST API, bypassing code redeployments.
  • Native support for MCP servers allows agents to interact with external tools using HTTP or stdio transports.
  • The platform retains full ADK compatibility, preserving standard sessions, events, and traces.
  • Built-in operational tools include token quotas, audit trails, and support for RAG and Text-to-SQL.

The Bottom Line

This is the infrastructure layer ADK was missing. By treating agents as data rather than code, apowerb unlocks the scalability needed for real-world team deployments.