AWS launches Web Search on Amazon Bedrock AgentCore with tens of billions of documents, no external API required

AWS launches managed web search for AI agents with proprietary index

Amazon Web Services launched Web Search on Amazon Bedrock AgentCore, a fully managed web search capability that eliminates the need for third-party search APIs when building AI agents. The service is now generally available.

Web Search on Bedrock AgentCore connects to agents through the Model Context Protocol (MCP), allowing agents to retrieve current information from the web without infrastructure overhead. The service operates entirely within AWS, with query traffic never leaving AWS infrastructure.

Purpose-built web index at scale

The service is backed by a proprietary web index maintained by Amazon that spans tens of billions of documents, according to AWS. The company updates the index continuously, reflecting new content within minutes of publication.

This differs from many competing solutions that wrap third-party search engines. AWS operates the index directly, giving it control over coverage, freshness, and data handling.

Key technical features

Web Search on Bedrock AgentCore includes three core components:

Knowledge graph for factual queries: The service maintains a built-in knowledge graph that grounds entities and relationships, providing high-confidence responses for factual questions about entities, roles, and dates.

Semantic snippet extraction: Rather than returning raw HTML, the tool performs semantic snippet extraction optimized for model context windows. It pulls relevant passages from each web page and returns them in a form that minimizes token usage on boilerplate content.

MCP-compatible integration: Agents discover the web search tool through a standard tools/list call and invoke it like other MCP tools, with no custom parsing required.

Privacy and deployment model

AWS emphasizes that queries remain within AWS infrastructure. The AgentCore Gateway authenticates to the web search connector using IAM service roles rather than requiring API keys or external credentials.

The service requires two IAM permissions: bedrock-agentcore:InvokeGateway and bedrock-agentcore:InvokeWebSearch. The web search resource ARN is owned by AWS (account = aws), with authorization enforced per invocation.

Setup and implementation

Developers add web search to agents by attaching a Web Search Tool target to an existing AgentCore Gateway using connectorId: "web-search". The Gateway handles schema management, parameter governance, endpoint resolution, and service authentication.

According to AWS documentation, setup requires:

• An AWS account with permissions to create IAM roles and Bedrock AgentCore resources
• AWS CLI v2 or Console access
• Python 3.10 or later for SDK integration
• An existing or new Amazon Bedrock AgentCore Gateway

The service is billable based on AgentCore Gateway usage and web search invocations. Pricing details were not disclosed in the announcement.

What this means

Web Search on Bedrock AgentCore addresses a core limitation of AI agents: knowledge cutoff dates. By building and maintaining its own web index rather than relying on third-party providers, AWS has eliminated API key management, quota concerns, and data egress issues that typically complicate web search integration. The minute-level index refresh rate and semantic snippet extraction suggest AWS is positioning this as an enterprise-grade alternative to solutions like Perplexity's API or wrapper services around Google/Bing. The MCP compatibility is notable—it signals AWS is building toward protocol standardization rather than proprietary lock-in, at least at the tool invocation layer.