ADK Scratchpad
Guides and sample agents for the Google Agent Development Kit (ADK) — from single-agent basics to multi-agent pipelines with cloud deployment.
ADK Scratchpad is a hands-on reference for building agents with the Google Agent Development Kit (ADK).
The adk_guide/ directory walks through ADK concepts step by step — environment setup, adding tools (Python functions, Google Search, MCP), and orchestrating multi-agent workflows with Sequential, Parallel, and Loop agents.
The agents/ directory contains standalone examples that range from simple single-agent setups to complex multi-stage pipelines:
- google_search — live web search via the built-in Google Search tool
- rag — retrieval-augmented generation backed by Vertex AI RAG Engine
- short_story — a SequentialAgent + LoopAgent pipeline that plans, writes, and iteratively edits stories
- bq_agent — BigQuery analysis with investigation sub-agents
- software-bug-assistant — combines PostgreSQL via MCP Toolbox, GitHub MCP, RAG, and LangChain tools
- travel-concierge — six sub-agents covering inspiration through post-trip review, with session state and Maps API
- ai-location-strategy — a seven-stage retail site-selection pipeline with Maps, code execution, and generated HTML reports
- short-movie-agents — end-to-end video creation using Imagen 4 for storyboards and Veo 3 for clips
Key patterns covered include custom function tools, MCP server integration, LangChain tool adapters, session and memory state, observability via OpenTelemetry, and deployment to Cloud Run and Vertex AI Agent Engine.