Skillget

Master-Agentic-AI-with-Google-ADK

mcp
Guvenlik Denetimi
Gecti
Health Gecti
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 23 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

Master Agentic AI with Google ADK: 117 runnable Python agents for the Agent Development Kit - tools, MCP, A2A, multi-agent workflows, sessions and memory, RAG, Live API voice, adk eval, tracing and Cloud Run deployment

README.md

Master Agentic AI with Google ADK

100+ ADK code samples covering every major ADK feature

Runnable Python agents for the Google Agent Development Kit: tools, MCP, A2A,
multi-agent workflows, sessions and memory, RAG, the Live API, evaluation and
deployment. Run any of them with adk web.

ADK
Python
Gemini
License
Stars
YouTube

About me

  • Vishal Bulbule, founder of TechTrapture
  • Google Developer Expert for Google Cloud and AI, enterprise architect
  • Contributor to the open-source Google ADK project
  • Teaching Agentic AI and MCP at TechTrapture Academy
  • 5 years of Google Cloud content on YouTube

This is the code behind the Master Agentic AI with Google ADK playlist,
rebuilt and tested on ADK 2.9.2 (google-genai 2.24, mcp 2.2, fastmcp 4.0,
a2a-sdk 1.1, Python 3.12, gemini-3.5-flash), verified 2026-09-22.

Topics covered

Google ADK tutorial, AI agents in Python, LlmAgent, function tools, OpenAPI
tools, Model Context Protocol (MCP) clients and servers, FastMCP, Agent2Agent
(A2A) protocol, sequential, parallel and loop workflows, Workflow graphs,
callbacks and plugins, guardrails, sessions, state and long-term memory,
artifacts, Google Search grounding, Vertex AI Search, agentic RAG, streaming
and the Live API for voice agents, adk eval, rubric metrics, OpenTelemetry
tracing, Cloud Run and Agent Runtime deployment, Secret Manager, CI/CD with
GitHub Actions.

Want to learn Agentic AI with guidance?

This repository is free and self-paced. For live sessions, real deployments
and feedback on your own agents, join the next
TechTrapture Academy cohort.

Join the next cohort

What is covered

  • Workflows: the ADK 2.x Workflow graph API for pipelines, fan-out,
    joins and loops, next to the SequentialAgent, ParallelAgent and
    LoopAgent classes it replaces
  • MCP, both directions: McpToolset clients, and servers built with
    FastMCP and the low-level mcp API
  • A2A 1.0: expose an agent with to_a2a, call it with RemoteA2aAgent
  • Callbacks at all six hook points, plus plugins, skills and guardrails
  • Memory: sessions, state prefixes, long-term memory, artifacts, rewind
  • Voice and streaming: the Live API, and token streaming
  • Evaluation: adk eval, pytest, rubric and custom metrics, OpenTelemetry
  • Deployment: Cloud Run and Agent Runtime, Secret Manager, CI/CD

Quick start

git clone https://github.com/vishal-bulbule/Master-Agentic-AI-with-Google-ADK.git
cd Master-Agentic-AI-with-Google-ADK

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

cp .env.example .env    # add a Gemini API key, or Agent Platform settings

SETUP.md covers accounts, keys, Google Cloud APIs, and the extra
setup a few modules need.

Running a sample

Every agent sample is a plain ADK agent folder (agent.py with a
root_agent, __init__.py, .env.example) with no runner code. Run it from
the folder that contains it:

cd Module_02_First_ADK_Agent
adk web                 # dev UI at http://localhost:8000

Open http://localhost:8000, pick the agent in the dropdown, and send a
prompt. The main panel's Events view lists what the agent did as numbered
rows (click one to see its details in the side panel), and the Traces view
next to it shows the timing of each model and tool call. The side panel's
State and Artifacts tabs show session state (read-only) and saved files.
adk run <agent_folder> runs the same agent in the terminal.

Each topic README follows the same layout: Prerequisites (extra packages,
keys, cloud resources to create first, servers that must be running), Run
it (the exact adk web command and flags), Try it (prompts to send and
what to look for), and Clean up. A few samples need flags such as
--session_service_uri or --a2a; their README gives the full command.

Every agent here has been run against a live model. The few that cannot be
verified end to end without extra setup, such as a microphone, a deployed
service or a third-party key, say so in their README.

The only things you run with python or pytest are the ones that are not
agents: the raw SDK scripts in Module 0, the MCP-without-ADK samples and MCP
servers in Modules 5 and 7, pytest suites, and the custom container app in
Module_15_Deployment/08_custom_dockerfile.

Modules

Work through them in order. Each module README lists its topics and the order
to follow.

# Module Agents What it covers
0 AI Foundations 1 The raw google-genai SDK underneath every agent: config, streaming, multimodal, function calling, tokens, MCP without ADK, then the same task as an ADK agent
1 Agents Landscape 5 The same task as a chatbot, a RAG pipeline and an agent; the agent loop traced
2 First ADK Agent 9 One agent in all three runtimes (adk web, adk run, adk api_server), YAML config, and five broken agents to debug
3 LlmAgent and Model Choice 11 Instructions, structured output, model routing, generation config, planners, LiteLLM
4 Custom Tools 12 Function tools, long-running tools, agents as tools, OpenAPI, auth, tool confirmation
5 MCP Fundamentals 1 MCP on its own: FastMCP servers, the JSON-RPC exchange, MCP Inspector, Claude Desktop, then an ADK agent as the host
6 MCP Client in ADK 9 McpToolset over stdio, SSE and Streamable HTTP; tool filters; GitHub and Google Maps servers
7 Build an MCP Server 1 FastMCP and low-level servers, exposing ADK tools over MCP, security, Cloud Run
8 Sessions, State and Memory 10 Session services, state prefixes, memory, artifacts, rewind
9 Workflows and Multi-Agent 12 Workflow graphs, parallel fan-out and join, loops, routing, custom agents, hierarchies
10 Callbacks, Plugins and Events 15 All six callback hooks, guardrails, caching, cost control, plugins, skills
11 Streaming and Live API 4 Token streaming, Live API voice agents in the dev UI, streaming tools
12 A2A Protocol 6 Exposing and consuming agents over A2A, agent cards, extensions
13 Grounding 7 Google Search grounding, citations, Vertex AI Search, agentic and hybrid RAG
14 Evaluation and Observability 8 Eval sets, criteria, pytest, rubrics, user simulation, tracing, safety layers
15 Deployment 6 Agent Runtime, Cloud Run, secrets, IAM, custom containers, scaling, CI/CD

Video walkthroughs

Every topic below has videos in the
Master Agentic AI with Google ADK playlist
on YouTube. Earlier videos were recorded on older ADK releases, so some APIs
and model names on screen have changed; the linked modules have the current,
tested code, and the code from the first videos is in
archived_samples/.

Getting started

Code: Module 1, Module 2

Tools

Code: Module 4

Agent types and workflows

Code: Module 9

Model Context Protocol (MCP)

Code: Module 5, Module 6, Module 7

Callbacks

Code: Module 10

Sessions, state and memory

Code: Module 8

Deployment

Code: Module 15

Real-world builds

Archived samples

The code from the original
YouTube playlist
is in archived_samples/. It is kept as recorded so it
matches the videos, and it is not maintained.

Author

Vishal Bulbule - Google Developer Expert for Google Cloud and AI, founder of
TechTrapture.

YouTube |
LinkedIn |
Medium |
X |
TechTrapture Academy

Issues and pull requests are welcome. If a sample breaks on a newer ADK
release, open an issue with the module path and the error.

License

Apache License 2.0. See LICENSE.

repo views

Yorumlar (0)

Sonuc bulunamadi