[#4442][feat] Add client-side A2A (Agent2Agent) protocol support to Scaffolding

examples/scaffolding/contrib/a2a/README.md

@@ -0,0 +1,57 @@
+# Scaffolding A2A (Agent2Agent) Example
+
+This example shows how a Scaffolding controller can delegate work to remote
+agents that speak the [A2A (Agent2Agent) protocol](https://a2a-protocol.org/),
+the agent-to-agent counterpart to the MCP tool-calling contrib. The generation
+model decides which remote agent to call; the `A2AWorker` forwards the message
+over A2A and feeds the reply back to the model for a final answer.
+
+This is a **client-side** integration: Scaffolding acts as an A2A client that
+consumes other agents. (Exposing a Scaffolding pipeline *as* an A2A server is a
+possible follow-up.)
+
+## Install
+
+```bash
+pip install a2a-sdk httpx uvicorn
+```
+
+`a2a-sdk` is only needed when actually talking to a remote agent; the contrib
+imports it lazily, and the unit tests do not require it.
+
+## Step 1: Start a remote A2A agent
+
+A minimal reference agent server is included:
+
+```bash
+python weather_agent_server.py --port 9999
+```
+
+This exposes a `weather_agent` whose agent card is discoverable at
+`http://0.0.0.0:9999/.well-known/agent-card.json`. You can also point the
+example at any other A2A-compatible agent server.
+
+## Step 2: Run the orchestrator
+
+```bash
+python a2a_run.py \
+    --API_KEY YOUR_API_KEY \
+    --base_url https://your-openai-compatible-endpoint/v1 \
+    --model your-model \
+    --agent_urls http://0.0.0.0:9999 \
+    --prompt "What is the weather in LA?"
+```
+
+The generation model receives the remote agents as callable tools, delegates to
+`weather_agent`, and summarizes its reply.
+
+## Files
+
+| File | Role |
+|------|------|
+| `a2a_run.py` | Scaffolding A2A orchestrator runner (client) |
+| `weather_agent_server.py` | Minimal reference A2A agent server for local testing |
+
+> `a2a-sdk` server/client APIs evolve across versions. The scripts target the
+> SDK's published "helloworld" pattern; adjust imports if your installed version
+> differs.

examples/scaffolding/contrib/a2a/a2a_run.py

@@ -0,0 +1,84 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Run a Scaffolding A2A orchestrator against one or more remote A2A agents.
+
+The generation side uses an OpenAI-compatible endpoint (any vendor, or a local
+``trtllm-serve``); the orchestration side talks the Agent2Agent protocol via
+``A2AWorker``. See README.md for how to start a sample remote agent server.
+"""
+
+import argparse
+import asyncio
+
+from openai import AsyncOpenAI
+
+from tensorrt_llm.scaffolding import OpenaiWorker, ScaffoldingLlm
+from tensorrt_llm.scaffolding.contrib.a2a import A2AController, A2AWorker
+from tensorrt_llm.scaffolding.contrib.mcp.chat_handler import chat_handler
+from tensorrt_llm.scaffolding.contrib.mcp.chat_task import ChatTask
+
+
+def parse_arguments():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--base_url",
+        type=str,
+        default="https://dashscope.aliyuncs.com/compatible-mode/v1",
+        help="OpenAI-compatible base URL for the generation model.",
+    )
+    parser.add_argument("--model", type=str, default="qwen-plus-latest")
+    parser.add_argument("--API_KEY", type=str)
+    parser.add_argument(
+        "--agent_urls",
+        type=str,
+        nargs="+",
+        default=["http://0.0.0.0:9999"],
+        help="Base URLs of the remote A2A agents to orchestrate.",
+    )
+    parser.add_argument("--prompt", type=str, default="What is the weather like today in LA?")
+    return parser.parse_args()
+
+
+async def main():
+    args = parse_arguments()
+
+    client = AsyncOpenAI(api_key=args.API_KEY, base_url=args.base_url)
+    generation_worker = OpenaiWorker(client, args.model)
+    generation_worker.register_task_handler(ChatTask, chat_handler)
+
+    a2a_worker = await A2AWorker.init_with_urls(args.agent_urls)
+
+    controller = A2AController()
+    llm = ScaffoldingLlm(
+        controller,
+        {
+            A2AController.WorkerTag.GENERATION: generation_worker,
+            A2AController.WorkerTag.A2A: a2a_worker,
+        },
+    )
+
+    future = llm.generate_async(args.prompt)
+    result = await future.aresult()
+    print(f"\nresult is {result.outputs[0].text}\n")
+
+    print("shutting down...")
+    llm.shutdown()
+    generation_worker.shutdown()
+    await a2a_worker.async_shutdown()
+    print("shut down done")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

examples/scaffolding/contrib/a2a/weather_agent_server.py

@@ -0,0 +1,87 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""A minimal reference A2A agent server used to exercise the A2A contrib.
+
+This follows the ``a2a-sdk`` "helloworld" server pattern and exposes a single
+``weather_agent`` that returns a canned reply. Run it with::
+
+    pip install a2a-sdk uvicorn
+    python weather_agent_server.py --port 9999
+
+then point ``a2a_run.py --agent_urls http://0.0.0.0:9999`` at it.
+
+Note: ``a2a-sdk`` server APIs evolve; this script targets the published
+helloworld example. Adjust imports if your installed SDK version differs.
+"""
+
+import argparse
+
+import uvicorn
+from a2a.server.agent_execution import AgentExecutor, RequestContext
+from a2a.server.apps import A2AStarletteApplication
+from a2a.server.events import EventQueue
+from a2a.server.request_handlers import DefaultRequestHandler
+from a2a.server.tasks import InMemoryTaskStore
+from a2a.types import AgentCapabilities, AgentCard, AgentSkill
+from a2a.utils import new_agent_text_message
+
+
+class WeatherAgentExecutor(AgentExecutor):
+    """Returns a canned weather reply regardless of the incoming message."""
+
+    async def execute(self, context: RequestContext, event_queue: EventQueue) -> None:
+        await event_queue.enqueue_event(new_agent_text_message("It is sunny in LA, around 75F."))
+
+    async def cancel(self, context: RequestContext, event_queue: EventQueue) -> None:
+        raise NotImplementedError("cancel is not supported by this agent")
+
+
+def build_app(host: str, port: int) -> A2AStarletteApplication:
+    skill = AgentSkill(
+        id="weather",
+        name="weather",
+        description="Returns the current weather for a location.",
+        tags=["weather"],
+        examples=["What is the weather in LA?"],
+    )
+    agent_card = AgentCard(
+        name="weather_agent",
+        description="A demo agent that reports the weather.",
+        url=f"http://{host}:{port}/",
+        version="1.0.0",
+        default_input_modes=["text"],
+        default_output_modes=["text"],
+        capabilities=AgentCapabilities(streaming=False),
+        skills=[skill],
+    )
+    request_handler = DefaultRequestHandler(
+        agent_executor=WeatherAgentExecutor(),
+        task_store=InMemoryTaskStore(),
+    )
+    return A2AStarletteApplication(agent_card=agent_card, http_handler=request_handler)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--host", default="0.0.0.0")
+    parser.add_argument("--port", type=int, default=9999)
+    args = parser.parse_args()
+
+    app = build_app(args.host, args.port)
+    uvicorn.run(app.build(), host=args.host, port=args.port)
+
+
+if __name__ == "__main__":
+    main()

tensorrt_llm/scaffolding/contrib/a2a/__init__.py

@@ -0,0 +1,27 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from .a2a_controller import A2AController
+from .a2a_task import A2AListTask, A2ASendTask
+from .a2a_utils import A2AAgentConnection, AgentInfo
+from .a2a_worker import A2AWorker
+
+__all__ = [
+    "A2AController",
+    "A2AWorker",
+    "A2ASendTask",
+    "A2AListTask",
+    "A2AAgentConnection",
+    "AgentInfo",
+]