使用自然語言查詢在 AgentCore 閘道中搜尋工具

如果您在建立閘道時啟用了閘道的語意搜尋，則可以呼叫 x_amz_bedrock_agentcore_search工具，使用自然語言查詢在閘道中搜尋工具。當您有許多工具且需要為您的使用案例尋找最合適的工具時，語意搜尋特別有用。若要了解如何在閘道建立期間啟用語意搜尋，請參閱建立 Amazon Bedrock AgentCore 閘道。

支援語意搜尋 AWS 的區域

下列 AWS 區域支援語意搜尋：

區域名稱	區域
美國東部 (維吉尼亞北部)	us-east-1
美國東部 (俄亥俄)	us-east-2
美國西部 (奧勒岡)	us-west-2
亞太地區 (海德拉巴)	ap-south-2
亞太區域 (孟買)	ap-south-1
亞太區域 (首爾)	ap-northeast-2
亞太區域 (新加坡)	ap-southeast-1
亞太區域 (雪梨)	ap-southeast-2
亞太區域 (東京)	ap-northeast-1
加拿大 (中部)	ca-central-1
歐洲 (法蘭克福)	eu-central-1
歐洲 (愛爾蘭)	eu-west-1
歐洲 (倫敦)	eu-west-2
歐洲 (米蘭)	eu-south-1
Europe (Paris)	eu-west-3
歐洲 (西班牙)	eu-south-2
Europe (Stockholm)	eu-north-1
南美洲 (聖保羅)	sa-east-1

若要使用此 AgentCore 工具搜尋工具，請使用 tools/call方法向閘道的 MCP 端點提出下列 POST 請求：

範例

注意

閘道僅接受其protocolConfiguration.mcp組態 supportedVersions 欄位中列出的 MCP 通訊協定版本。若要使用版本 2026-07-28，請確定閘道的 supportedVersions包含它。您可以使用 UpdateGateway API 變更支援的版本。

取代以下的值：

${GatewayEndpoint} – 閘道的 URL，如 CreateGateway API 的回應所提供。
${Authorization header} – 當您設定傳入授權時，來自身分提供者的授權憑證。
${McpProtocolVersion} – 請求的 MCP 通訊協定版本，例如 2025-11-25。版本必須是閘道支援的版本。
${RequestName} – 請求的名稱。
${Query} – 搜尋工具的自然語言查詢。

回應會傳回與查詢相關的工具清單。

工具搜尋的程式碼範例

若要查看使用自然語言查詢在閘道中尋找工具的範例，請選取下列其中一種方法：

範例

Python requests package (2025-11-25 and earlier)

將 MCP-Protocol-Version標頭設定為閘道支援的版本。


import requests
import json

def search_tools(gateway_url, access_token, query):
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {access_token}",
        "MCP-Protocol-Version": "2025-11-25"
    }

    payload = {
        "jsonrpc": "2.0",
        "id": "search-tools-request",
        "method": "tools/call",
        "params": {
            "name": "x_amz_bedrock_agentcore_search",
            "arguments": {
                "query": query
            }
        }
    }

    response = requests.post(gateway_url, headers=headers, json=payload)
    return response.json()

# Example usage
gateway_url = "https://${GatewayEndpoint}/mcp" # Replace with your actual gateway endpoint
access_token = "${AccessToken}" # Replace with your actual access token
results = search_tools(gateway_url, access_token, "find order information")
print(json.dumps(results, indent=2))

Python requests package (2026-07-28)

在版本上2026-07-28，在內文中包含 Mcp-Method和 Mcp-Name request-metadata 標頭和_meta版本欄位。MCP-Protocol-Version 標頭必須符合 _meta.io.modelcontextprotocol/protocolVersion。閘道的 supportedVersions 必須包含 2026-07-28。


import requests
import json

def search_tools(gateway_url, access_token, query):
    headers = {
        "Content-Type": "application/json",
        "Authorization": f"Bearer {access_token}",
        "MCP-Protocol-Version": "2026-07-28",
        "Mcp-Method": "tools/call",
        "Mcp-Name": "x_amz_bedrock_agentcore_search"
    }

    payload = {
        "jsonrpc": "2.0",
        "id": "search-tools-request",
        "method": "tools/call",
        "params": {
            "name": "x_amz_bedrock_agentcore_search",
            "arguments": {
                "query": query
            },
            "_meta": {
                "io.modelcontextprotocol/protocolVersion": "2026-07-28",
                "io.modelcontextprotocol/clientInfo": {"name": "my-agent", "version": "1.0.0"},
                "io.modelcontextprotocol/clientCapabilities": {}
            }
        }
    }

    response = requests.post(gateway_url, headers=headers, json=payload)
    return response.json()

# Example usage
gateway_url = "https://${GatewayEndpoint}/mcp" # Replace with your actual gateway endpoint
access_token = "${AccessToken}" # Replace with your actual access token
results = search_tools(gateway_url, access_token, "find order information")
print(json.dumps(results, indent=2))

MCP Client


from mcp import ClientSession
from mcp.client.streamable_http import streamablehttp_client
import asyncio

async def execute_mcp(
    url,
    token,
    tool_params,
    headers=None
):
    default_headers = {
        "Authorization": f"Bearer {token}"
    }
    headers = {**default_headers, **(headers or {})}

    async with streamablehttp_client(
       url=url,
       headers=headers,
    ) as (
        read_stream,
        write_stream,
        callA,
    ):
        async with ClientSession(read_stream, write_stream) as session:
            # 1. Perform initialization handshake
            print("Initializing MCP...")
            _init_response = await session.initialize()
            print(f"MCP Server Initialize successful! - {_init_response}")

            # 2. Call specific tool
            print(f"Calling tool: {tool_params['name']}")
            tool_response = await session.call_tool(
                name=tool_params['name'],
                arguments=tool_params['arguments']
            )
            print(f"Tool response: {tool_response}")
            return tool_response

async def main():
    url = "https://${GatewayEndpoint}/mcp"
    token = "your_bearer_token_here"
    tool_params = {
        "name": "x_amz_bedrock_agentcore_search",
        "arguments": {
            "query": "How do I find order details?"
        }
    }
    await execute_mcp(
        url=url,
        token=token,
        tool_params=tool_params
    )


if __name__ == "__main__":
    asyncio.run(main())

Strands MCP Client


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

def create_streamable_http_transport(mcp_url: str, access_token: str):
    return streamablehttp_client(mcp_url, headers={"Authorization": f"Bearer {access_token}"})

def get_full_tools_list(client):
    """
    List tools w/ support for pagination
    """
    more_tools = True
    tools = []
    pagination_token = None
    while more_tools:
        tmp_tools = client.list_tools_sync(pagination_token=pagination_token)
        tools.extend(tmp_tools)
        if tmp_tools.pagination_token is None:
            more_tools = False
        else:
            more_tools = True
            pagination_token = tmp_tools.pagination_token
    return tools

def run_agent(mcp_url: str, access_token: str):
    mcp_client = MCPClient(lambda: create_streamable_http_transport(mcp_url, access_token))

    with mcp_client:
        tools = get_full_tools_list(mcp_client)
        print(f"Found the following tools: {[tool.tool_name for tool in tools]}")
        result = mcp_client.call_tool_sync(
            tool_use_id="tool-123",  # A unique ID for the tool call
            name="x_amz_bedrock_agentcore_search",  # The name of the tool to invoke
            arguments={"query": "find order information"}  # A dictionary of arguments for the tool
        )
        print(result)

url = {gatewayUrl}
token = {AccessToken}
run_agent(url, token)

LangGraph MCP Client


import asyncio

from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent

url = ""
headers = {}

def filter_search_tool(
):
    mcp_client = MultiServerMCPClient(
        {
            "agent": {
                "transport": "streamable_http",
                "url": url,
                "headers": headers,
            }
        }
    )
    tools = asyncio.run(mcp_client.get_tools())
    builtin_search_tool = []
    for tool in tools:
        if tool.name == "x_amz_bedrock_agentcore_search":
            builtin_search_tool.append(tool)
    return builtin_search_tool

def execute_agent(
    user_prompt,
    model_id,
    region,
    tools
):
    model = ChatBedrock(model_id=model_id, region_name=region)

    agent = create_react_agent(model, filter_search_tool())
    _response = asyncio.run(agent.ainvoke({
        "messages": user_prompt
    }))

    _response = _response.get('messages', {})[1].content
    print(
        f"Invoke Langchain Agents Response"
        f"Response - \n{_response}\n"
    )
    return _response

您的瀏覽器已停用或無法使用 Javascript。

您必須啟用 Javascript，才能使用 AWS 文件。請參閱您的瀏覽器說明頁以取得說明。

文件慣用形式

抽樣

連接代理程式