ADK agentic patterns (Memory + MCP)

Hands-on Google ADK tutorial covering multi-agent workflows for trip planning.

What I learned

Pattern	Folder	Example prompt
Sequential	b1_sequential_agent	Find the best sushi restaurant in Bangalore, then directions from MG Road.
Parallel	b2_parallel_agent	Suggest a museum, a concert, and a restaurant for a weekend in Bangalore.
Loop	b3_loop_agent	Plan one activity + one restaurant in SF with travel time under 45 minutes.
Custom	c_custom_agent	I have a $100 budget—build a day plan in Sunnyvale.
Routing	d_routing_agent	I'm hungry for Italian near Indiranagar.
Agent as tool	e_agent_as_tool	Plan a museum + dinner in Sunnyvale and validate logistics.
Single agent	a_single_agent	Weekend date plan in Bangalore, foodie + outdoors, July 19–20.
Memory	f_agent_with_memory	Remember I like Italian food, then plan something in Sunnyvale.
MCP tools	g_agents_mcp	Find top-rated museums in Tokyo under $20.

Local setup (short)

Prereqs: Python 3.12+, Ollama with llama3.2

git clone https://github.com/Nithya113/adk-agentic-pattern-memory-mcp.git
cd adk-agentic-pattern-memory-mcp

# Prefer Homebrew Python on Mac (avoids Anaconda openssl issues)
/opt/homebrew/bin/python3.12 -m venv .adk_env
source .adk_env/bin/activate
pip install -U pip
pip install -r requirements.txt
pip install litellm

# Optional: Gemini API (not needed if using Ollama)
# cat > .env <<'EOF'
# GOOGLE_GENAI_USE_VERTEXAI=FALSE
# GOOGLE_API_KEY=your_key
# EOF

# Ollama (OpenAI-compatible)
cat >> .env <<'EOF'
OLLAMA_API_BASE=http://localhost:11434
OPENAI_API_BASE=http://localhost:11434/v1
OPENAI_API_KEY=ollama
EOF

ollama pull llama3.2
adk web   # http://127.0.0.1:8000

MCP trip agent (optional)

python setup_trip_database.py

# Terminal 1 — download toolbox for your OS from genai-toolbox, then:
cd mcp_tool_box
chmod +x toolbox
./toolbox --tools-file trip_tools.yaml --port 7001

# Terminal 2
cd g_agents_mcp && python main.py
# or use g_agents_mcp in ADK web while toolbox is running

Original course: goo.gle/advancedadk