ToolExperimentalMCP

adk-web-multi-agent

Name: adk-web-multi-agent
Rating: 3.0 (38 reviews)
Author: seehiong

by seehiong

Notebook demo of ADK-based multi-agent coordination querying Postgres and Data Commons

Jupyter Notebook

Updated Nov 3, 2025

Stars

Forks

View on GitHub

Overview

Coordinates specialist agents built with Google ADK to query PostgreSQL and Data Commons using OpenRouter models. Implements a coordinator pattern (director agent) that routes subtasks to specialist agents and aggregates responses using the MCP Toolbox. The Jupyter notebook demonstrates end-to-end flows for HDB resale lookups and global statistics, making the architecture easy to reproduce and adapt. Model Context Protocol (MCP).

Key Benefits

As agents delegate work to specialists, understanding who handled which subtask and how results were produced becomes crucial for trust and troubleshooting. This repo provides a concrete example of agent delegation, data access, and aggregation so practitioners can inspect interaction patterns and spot failure modes. Until now most examples focused on single-agent apps — this shows practical multi-agent orchestration tied to real data sources, useful for building agent-to-agent evaluation and reputational signals. See how these patterns relate to Agent Registry Pattern and Semantic Capability Matching Pattern.

Target Use Cases

Researchers and engineers prototyping multi-agent workflows that need concrete examples of delegation, MCP-based messaging, and data-backed agents. For a structured approach, explore Evaluation-Driven Development Pattern.

Applications

Prototype coordinator-directed agent pipelines that query relational and public data sources
Inspect agent interaction patterns to identify delegation and aggregation failure modes
Build reproducible examples for agent-to-agent evaluation and reputation signal design

Test in the Playground

Validate against scenarios