🤖 A2A Protocol

What is A2A?

Instead of one monolithic model doing everything, the A2A pattern splits work across focused agents that expose their capabilities and talk to each other over plain HTTP. A coordinator analyzes each problem, routes it to the right specialist agent(s), and merges the results — and because the transport is just JSON-RPC, agents built on different frameworks can interoperate.

Architecture at a glance

graph TD
    U["👤 User / Caller"] --> O["🧭 Multi-Agent Orchestrator
analyze · route · combine"]
    O -->|JSON-RPC 2.0| MA["🧮 Math Agent
:8001"]
    O -->|JSON-RPC 2.0| DA["📊 Data Analyst Agent
:8002"]
    MA <-->|A2A Protocol over HTTP| DA
        

Every agent publishes an Agent Card at / and /.well-known/agent.json for runtime capability discovery. Read the full design — sequence diagrams, data model and extension guide — in the Technical Architecture.

Quick start

# 1. Clone & install
git clone https://github.com/rrahimi-uci/a2a-poc.git
cd a2a-poc
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# 2. Launch the whole system (both agents + demo)
python scripts/launch_system.py

Or talk to an agent directly with one curl:

curl -X POST http://localhost:8001/ \
  -H "Content-Type: application/json" \
  -d '{"id":"1","jsonrpc":"2.0","method":"message/send",
       "params":{"message":{"role":"user",
       "parts":[{"kind":"text","text":"calculate mean of [1,2,3,4,5]"}]}}}'

Why it's useful

• Interoperable — plain HTTP + JSON-RPC 2.0; any language or framework can join.
• Discoverable — Agent Cards let agents learn each other's capabilities at runtime.
• Composable — an orchestrator combines specialists into richer solutions.
• Cross-framework — includes a hybrid demo with a Microsoft AutoGen-style agent.
• Tested — a fast, fully in-process pytest suite (31 tests) keeps it honest.

Learn more