Skip to content

README.md

Latest commit

 

History

History
114 lines (87 loc) · 4.18 KB

README.md

File metadata and controls

114 lines (87 loc) · 4.18 KB
Raw

A2A ↔ ADK Interop Example

This example shows how the trpc-agent-go A2A client talks to an ADK Python A2A server.

Two example cases are provided:

Case Server Client Features
Tool Call adk/adk_server.py (port 8081) trpc_agent/main.go Streaming + tools (calculator, current_time)
Code Execution adk/adk_codeexec_server.py (port 8082) trpc_agent_codeexec/main.go Streaming + Python code execution

Directory layout

examples/a2aadk/
├── README.md                       # This guide
├── trpc_agent/
│   └── main.go                     # Go client for tool call example
├── trpc_agent_codeexec/
│   └── main.go                     # Go client for code execution example
└── adk/
    ├── adk_server.py               # ADK server with tools
    └── adk_codeexec_server.py      # ADK server with code execution

Prerequisites

Component Version Notes
Go 1.21+ Needed for trpc_agent/main.go
Python 3.10+ (3.11 recommended) Required by the ADK server
Pip deps See adk/requirements.txt Install via pip install -r requirements.txt
API key OPENAI_API_KEY Example server defaults to OpenAI-compatible models

Prepare the ADK/Python side

  1. (Optional) Create a virtualenv: 
    cd trpc-agent-go/examples/a2aadk/adk
python3 -m venv venv
source venv/bin/activate
  2. Install dependencies: 
    pip install -r requirements.txt
  3. Configure model access: 
    export OPENAI_API_KEY="<your-openai-key>"
# Optional overrides
export MODEL_NAME="gpt-4o-mini"
export OPENAI_API_URL="https://your-endpoint/v1"

Case 1: Tool Call Example

Start the server

cd trpc-agent-go/examples/a2aadk/adk
python3 adk_server.py

Run the client

cd trpc-agent-go
go run ./examples/a2aadk/trpc_agent --url http://localhost:8081

Server highlights:

  • Streaming responses plus two example tools: calculator and current_time.
  • logging_request_converter prints user_id / session_id for debugging.

Case 2: Code Execution Example

Start the server

cd trpc-agent-go/examples/a2aadk/adk
python3 adk_codeexec_server.py

Run the client

cd trpc-agent-go
go run ./examples/a2aadk/trpc_agent_codeexec --url http://localhost:8082

Server highlights:

  • ADK agent with PythonCodeExecution enabled.
  • LLM can generate and execute Python code, returning results via A2A protocol.

Client behavior:

  • Displays code execution events (💻 Code Execution) and results (📊 Execution Result) separately.
  • Uses MessageID-based deduplication to handle ADK's repeated event delivery.

ADK-specific notes

Scenario Details
Cumulative streaming ADK A2A events send "full content so far" in every delta. The client captures the last valid intermediate event rather than reading deltas incrementally.
Final event duplication ADK may send a malformed final event that duplicates content. The client ignores the final event payload.
Code execution event duplication ADK may send repeated code execution events with the same MessageID. The client uses MessageID-based deduplication.
User propagation a2aagent places Session.UserID into the X-User-ID header. Override via a2aagent.WithUserIDHeader(...).

FAQ

  1. "OPENAI_API_KEY not set" warning: the server cannot call OpenAI without the env var. Run export OPENAI_API_KEY=... before starting adk_server.py.
  2. Go client cannot connect: ensure the ADK server is listening on the --url you pass (default http://localhost:8081). Adjust either the server port or the flag.
  3. Need to disable ADK compatibility? This example assumes an ADK peer, so trpc-agent-go keeps the adk_ metadata prefix enabled. If you target a non-ADK service you can call a2a.WithADKCompatibility(false) in your own server, but no change is required here.

After completing the steps above, the Go client and ADK server will interoperate and you can observe tool invocations plus final answers directly in the terminal. Happy debugging!