Bedrock AgentCore Gains Zero-Egress Web Search via MCP Gateway

On June 17, 2026, AWS announced the general availability of Web Search on Amazon Bedrock AgentCore. Unveiled at the AWS Summit in New York, the fully managed capability allows AI agents to securely query real-time public web data without relying on external search APIs. This solves the knowledge freshness problem for production models while maintaining strict enterprise data residency policies.

Architecture and Search Infrastructure

The service is implemented as a built-in connector target on the Bedrock AgentCore Gateway. AWS exposes this functionality through the Model Context Protocol (MCP), creating a standardized integration path for agentic frameworks such as LangChain, LangGraph, and Strands. Developers can route retrieval requests through the gateway without writing custom glue code or managing third-party authentication tokens.

Search queries execute against Amazon’s proprietary index, the same backend infrastructure utilized by Alexa+, Amazon Quick Suite, and Kiro. This index spans tens of billions of documents and refreshes within minutes of new content publication. The tool also queries Amazon’s Knowledge Graph to pull structured entity data and verified real-time facts, including stock prices and sports scores, directly into the agent context.

Privacy and Payload Formatting

Enterprise adoption of web-grounded agents often stalls on data governance. The Bedrock AgentCore implementation guarantees zero data egress to third-party search providers. All prompt data and retrieval queries remain strictly within the customer’s secured AWS environment. Early adopters like Sony and Nasdaq integrated the capability to establish a consistent governance model across their AI deployments, eliminating the risk of leaking sensitive inputs to external search engines.

When an agent initiates a search, the gateway returns a ranked JSON payload designed for direct prompt injection.

Response Component	Description
Semantic Snippets	Contextual text chunks relevant to the specific query.
Source URLs	Direct links to the origin domain for verification.
Metadata	Page titles and publication dates to establish timeline context.

This structured response format allows agents to generate deterministic citations and rank conflicting information based on publication recency.

Pricing and Service Quotas

AWS is rolling out the feature initially in the US East (N. Virginia) region. The service operates on a flat pricing model of $7.00 per 1,000 queries.

To support production scaling, AWS simultaneously increased AgentCore Runtime service quotas. Active session workloads per account are now capped at 5,000 in the US East (N. Virginia) and US West (Oregon) regions. The InvokeAgentRuntime API rate limit also increased from 25 to 200 transactions per second (TPS).

Expanded Agentic Capabilities

The Web Search launch arrived alongside a broader suite of updates to the Bedrock agent stack. AWS introduced the Amazon Bedrock Managed Knowledge Base, which handles multi-step internal data queries via an “Agentic Retriever” and includes smart parsing for complex documents.

Security at the gateway level also expanded with the AgentCore Policy with Bedrock Guardrails. This feature evaluates tool inputs and agent outputs for prompt injections and sensitive data exposure before the payload reaches the model. Concurrently, content owners gained the ability to charge bots for access via the new AWS WAF AI Traffic Monetization feature, establishing a technical foundation for a pay-per-intelligence data ecosystem.

If you currently maintain custom search integrations for your AWS-hosted agents, evaluate migrating to the native gateway connector. Standardizing on the built-in MCP target removes the operational overhead of key rotation and third-party rate limits while bringing web retrieval inside your existing compliance boundary.