I tried the Amazon Bedrock AgentCore Web Search Tool, which became available in the Tokyo region, with Japanese queries
This page has been translated by machine translation. View original
Introduction
On August 19, 2026, AWS What's New announced support for domain filters and published date filters in Amazon Bedrock AgentCore Web Search Tool, along with an expansion of available regions. In addition to the existing US East (N. Virginia), Europe (Ireland) and Asia Pacific (Tokyo) were added as available regions.
Web Search Tool is a search tool available from MCP clients via AgentCore Gateway.
In this article, in addition to searching with queries containing Japanese, we verified the index registration rate of DevelopersIO articles and the lag for new article registration.
Verification Details
We created a Gateway and a GatewayTarget for the Web Search Tool using CloudFormation, and executed searches from an MCP client implemented in Python.
Verification Environment
| Item | Value |
|---|---|
| Region | ap-northeast-1 |
| Gateway Name | web-search-verify-<random suffix> |
| GatewayTarget | web-search-tool (connector version: 1.1.0) |
| Authentication Method | AuthorizerType: AWS_IAM (no Cognito/OAuth required) |
Environment Setup and MCP Connectivity
We created a full set of resources including the Gateway execution role using CloudFormation. The main resource definitions are as follows.
WebSearchGateway:
Type: AWS::BedrockAgentCore::Gateway
Properties:
Name: !Ref GatewayName
Description: Web Search Tool動作検証用のAgentCore Gateway
RoleArn: !GetAtt GatewayExecutionRole.Arn
ProtocolType: MCP
AuthorizerType: AWS_IAM
WebSearchGatewayTarget:
Type: AWS::BedrockAgentCore::GatewayTarget
Properties:
GatewayIdentifier: !GetAtt WebSearchGateway.GatewayIdentifier
Name: web-search-tool
Description: Web Search Tool connector target
TargetConfiguration:
Mcp:
Connector:
Source:
ConnectorId: web-search
Configurations:
- Name: WebSearch
ParameterValues: {}
CredentialProviderConfigurations:
- CredentialProviderType: GATEWAY_IAM_ROLE
We created the Gateway with AuthorizerType: AWS_IAM and authenticated using SigV4 signing with temporary credentials obtained via AssumeRole. The following Python code POSTs an MCP tools/list request to the Gateway URL.
import botocore.session
from botocore.awsrequest import AWSRequest
from botocore.auth import SigV4Auth
session = botocore.session.get_session()
credentials = session.get_credentials() # Obtained from environment variables such as AWS_ACCESS_KEY_ID
request = AWSRequest(
method="POST",
url=GATEWAY_URL,
data=json.dumps(payload).encode("utf-8"),
headers={"Content-Type": "application/json", "Accept": "application/json, text/event-stream"},
)
SigV4Auth(credentials, "bedrock-agentcore", REGION).add_auth(request)
Calling tools/list returned the following response.
{"jsonrpc":"2.0","id":1,"result":{"tools":[{"inputSchema":{"type":"object","properties":{"maxResults":{"description":"Maximum number of results to return. Valid range: 1-25. Defaults to 10.","type":"integer"},"query":{"description":"The search query string","type":"string"}},"required":["query"]},"name":"web-search-tool___WebSearch"}]}}
The published tool name was web-search-tool___WebSearch. This is a format that concatenates the GatewayTarget name (web-search-tool) and the connector configuration name (WebSearch) with ___.
Searching with Japanese Queries
We tested whether DevelopersIO articles could be found using queries containing Japanese. The top 5 results returned for the query "Reproducing the nginx vulnerability (CVE-2026-42533) reported in July 2026 with Docker" were formatted for article publication. Rankings were assigned based on the order in the results array, and text excerpts are values truncated to the first 20 characters of the response body.
{
"results": [
{
"rank": 1,
"title": "2026年7月報告のnginx脆弱性（CVE-2026-42533）をDockerで再現してみた",
"url": "https://dev.classmethod.jp/articles/cve-2026-42533-nginx-rce-docker-repro/",
"publishedDate": "unknown",
"text_excerpt": "--- 2026年7月報告のnginx脆…"
},
{
"rank": 2,
"title": "CVE-2026-42533 · My images",
"url": "https://scout.docker.com/vulnerabilities/id/CVE-2026-42533/my-images",
"publishedDate": "unknown",
"text_excerpt": "--- CVE-2026-42533 A…"
},
{
"rank": 3,
"title": "CVE-2026-42533",
"url": "https://scout.docker.com/v/CVE-2026-42533?s=debian",
"publishedDate": "09:03PM, Sunday, July 19 2026, PDT",
"text_excerpt": "--- CVE-2026-42533 A…"
},
{
"rank": 4,
"title": "Developers.IO[B!]新着記事・評価",
"url": "https://b.hatena.ne.jp/entry/s/dev.classmethod.jp/articles/monitor-dx-connection-by-cloudwatch-alarm/",
"publishedDate": "unknown",
"text_excerpt": "- 2026年7月報告のnginx脆弱性…"
},
{
"rank": 5,
"title": "https://nginx.org/",
"url": "https://nginx.org/",
"publishedDate": "11:34AM, Friday, August 21 2026, PDT",
"text_excerpt": "| event_date | 2026-…"
}
]
}
On the other hand, here are the top 5 results returned for the query "Trying to bypass CloudFront VPC origin failures using Inter-Region VPC peering."
{
"results": [
{
"rank": 1,
"title": "I tried bypassing CloudFront VPC origin failures using Inter-Region VPC peering",
"url": "https://dev.classmethod.jp/en/articles/cloudfront-vpc-origin-failure-inter-region-vpc-peering-bypass/",
"publishedDate": "05:00PM, Wednesday, July 15 2026, PDT",
"text_excerpt": "--- I tried bypassin…"
},
{
"rank": 2,
"title": "CloudFront VPC オリジンで実現するマルチリージョンのアクティブ/アクティブ構成",
"url": "https://aws.amazon.com/jp/blogs/news/multi-region-active-active-architecture-with-cloudfront-vpc-origins/",
"publishedDate": "04:00PM, Wednesday, January 21 2026, PST",
"text_excerpt": "Amazon Web Services …"
},
{
"rank": 3,
"title": "CloudFront オリジンフェイルオーバーを使用して高可用性を最適化する",
"url": "https://docs.aws.amazon.com/ja_jp/AmazonCloudFront/latest/DeveloperGuide/high_availability_origin_failover.html",
"publishedDate": "unknown",
"text_excerpt": "CloudFront オリジンフェイルオ…"
},
{
"rank": 4,
"title": "Amazon CloudFront VPC オリジンの紹介: アプリケーションのセキュリティ強化と運用の合理化",
"url": "https://aws.amazon.com/jp/blogs/news/introducing-amazon-cloudfront-vpc-origins-enhanced-security-and-streamlined-operations-for-your-applications/",
"publishedDate": "04:00PM, Saturday, November 23 2024, PST",
"text_excerpt": "Amazon CloudFront 仮想…"
},
{
"rank": 5,
"title": "CloudFront Distribution Patterns for Multi-Region Failover",
"url": "https://blog.blazingcdn.com/en-us/cloudfront-distribution-patterns-multi-region-failover",
"publishedDate": "03:33PM, Thursday, May 28 2026, PDT",
"text_excerpt": "CloudFront Multi-Reg…"
}
]
}
The top result was the automatically translated English version of the target article. Even in this search where the Japanese title was specified as-is, the English version was displayed.
Index Coverage Rate
After confirming that searches with queries containing Japanese worked, we expanded the scope and investigated how well DevelopersIO articles are indexed in the search index. We specified 99 article URLs that had high click counts in Google Search Console as queries and searched with maxResults=5. A hit was determined when the target URL, or a URL for a different language version of the same article (such as /en/articles/), was included in the search results.
| Rank Range | Hits/Count | Hit Rate |
|---|---|---|
| Rank 1-20 | 17/20 | 85% |
| Rank 21-80 | 54/60 | 90% |
| Rank 81-99 | 13/19 | 68% |
| Total | 84/99 | 85% |
Overall, 85% (84/99) of the top-clicked articles were included in the AgentCore Web Search Tool search results. The 81–99 rank group had a 68% rate, which was lower than the other groups. However, the targets are articles with high click counts that have been published for a sufficient amount of time. Based on this data alone, it is not possible to determine whether article popularity or time elapsed since publication had more impact on the hit rate.
For the 15 articles that were not hits, it was notable that the English version URL of the target article or similar articles by the same author or on the same topic appeared in first place.
New Article Indexing Lag
We extracted 39 articles published in the past 8–9 days from the DevelopersIO sitemap and tabulated the hit rate by time elapsed since publication.
| Time Elapsed | Hits/Count | Hit Rate |
|---|---|---|
| 0-24h | 0/1 | 0% |
| 24-48h | 1/4 | 25% |
| 48-72h | 1/5 | 20% |
| 72-96h | 2/6 | 33% |
| 96-120h | 2/3 | 67% |
| 120-144h | 4/5 | 80% |
| 144-168h | 4/5 | 80% |
| 168-192h | 5/6 | 83% |
| 192-216h | 3/4 | 75% |
In this sample, articles published within 24 hours were not hits. The hit rate increased over time, reaching the 60–80% range 4–5 days after publication.
Updating to Connector 1.2.0 and Verifying Filter Behavior
The connector version used in the verification environment was 1.1.0, but it was possible to update to 1.2.0 using the AWS CLI's update-gateway-target.
aws bedrock-agentcore-control update-gateway-target \
--gateway-identifier <Gateway ID> \
--target-id <Target ID> \
--name web-search-tool \
--target-configuration '{"mcp":{"connector":{"source":{"connectorId":"web-search","version":"1.2.0"},"configurations":[{"name":"WebSearch","parameterValues":{}}]}}}' \
--credential-provider-configurations '[{"credentialProviderType":"GATEWAY_IAM_ROLE"}]'
After the update, parameters for domain filters and published date filters were added to the tool definition response.
{"filters":{"description":"Per-call runtime filters set by the caller for this specific request. Contains an optional runtime domainFilter (merged with the admin-level domainFilter) and an optional publishedDateFilter.","type":"object","properties":{"domainFilter":{"description":"Per-call domain filter. Include and exclude lists each hold up to 100 domains. Merged with the admin-level domainFilter.","type":"object","properties":{"include":{"type":"array","items":{"type":"string"}},"exclude":{"type":"array","items":{"type":"string"}}}},"publishedDateFilter":{"description":"Per-call published-date filter. Both bounds inclusive, UTC, ISO-8601. Applies to web results only.","type":"object","properties":{"from":{"type":"string"},"to":{"type":"string"}}}}}}
We searched with dev.classmethod.jp specified in the domain filter. All 5 returned results were articles from that domain, including the English version under /en/articles/.
{"filters": {"domainFilter": {"include": ["dev.classmethod.jp"]}}}
When a date range was specified in the published date filter, the publication dates of the search results were also within the specified range.
{"filters": {"publishedDateFilter": {"from": "2026-08-01", "to": "2026-08-22"}}}
Pricing
According to the AgentCore pricing page at the time of writing (August 2026), the search fee for Web Search Tool is $0.007 per search. Tool invocations (InvokeTool) on the Gateway side incur an additional $0.005 per 1,000 calls. We targeted a total of 139 calls for cost estimation: 1 for Japanese query operation verification, 99 for coverage investigation, and 39 for new article investigation. The search fee is 139 calls × $0.007 = $0.973, the InvokeTool fee is 139 calls × $0.005 / 1,000 ≈ $0.0007, for a total of approximately $0.97.
Summary
We tested searching with queries containing Japanese using Amazon Bedrock AgentCore Web Search Tool, which has become available in the Tokyo region. We confirmed that DevelopersIO articles, including English versions of the same articles, appear in search results at a high rate.
Please give it a try as one option for incorporating Japanese web search into AgentCore.
In addition to the domain and published date filters supported this time, we look forward to future expansion of filters that can narrow down search results by conditions such as language. We also hope to see the provision of a mechanism equivalent to Search Console that allows content providers to check and manage their search index registration status.
Reference Links
