strands-agents

Strands Agents

The default agent framework for this workspace. Strands is AWS's lightweight, code-first agent SDK: Bedrock is the default model provider, with Anthropic, OpenAI, and Gemini also supported, and it runs on AWS, on other clouds, or on-prem.

When to Use

• Writing a new agent or chatbot in this workspace
• Adding tools to an existing Strands agent
• Connecting an agent to an MCP server or an AgentCore Gateway
• Building supervisor/worker multi-agent flows

The Model-Driven Tool Loop

Strands does not ask you to draw a graph. You give the model a system prompt and a set of tools; the model decides the sequence.

from strands import Agent, tool

@tool
def lookup_order(order_id: str) -> dict:
    """Return status, items, and delivery estimate for a customer order.

    Args:
        order_id: The customer-facing order id, e.g. "ORD-10432".
    """
    return orders.get(order_id, {"error": "order not found", "retryable": False})

agent = Agent(
    model="anthropic.claude-sonnet-4-5-20250929-v1:0",
    system_prompt="You are a support assistant. Use tools before answering "
                  "questions about orders. If a tool returns an error, explain "
                  "it plainly; never invent an order status.",
    tools=[lookup_order],
)

result = agent("Where is ORD-10432?")

The docstring is the tool description the model sees. Write it for the model: what it does, what the arguments mean, and when not to use it.

Streaming

Perceived latency is the product's latency. Stream by default in anything user-facing.

async for event in agent.stream_async(prompt):
    if "data" in event:
        yield event["data"]

MCP Tools

Point the agent at a Gateway endpoint or a standalone MCP server and its tools join the loop alongside native @tool functions:

from mcp.client.streamable_http import streamablehttp_client
from strands.tools.mcp import MCPClient

gateway = MCPClient(lambda: streamablehttp_client(
    GATEWAY_URL, headers={"Authorization": f"Bearer {access_token}"}))

with gateway:
    agent = Agent(tools=[lookup_order, *gateway.list_tools_sync()])
    result = agent(prompt)