Amazon Launches Web Search API for AI Agents Inside AWS: $7/1K Queries, Zero Data Egress
Cloud Platform EngineeringEvery AI agent operating on training data alone fails on questions that depend on events, prices, research, or regulations that postdate its cutoff. AgentCore Web Search closes that gap inside AWS — without routing enterprise queries through third-party search providers.
What Happened
Amazon has launched Web Search on Amazon Bedrock AgentCore, generally available in US East (N. Virginia) as of June 17, 2026. The service gives AI agents access to current web knowledge via a managed API built on Amazon’s search infrastructure — the same technology underlying Alexa+, Amazon Quick, and Kiro.
Pricing is $7 per 1,000 queries. New AWS customers receive up to $200 in Free Tier credits.
What It Does
AgentCore Web Search returns snippets, source URLs, titles, and publication dates for each query — the structured context an agent needs to reason over current information before generating a response. Retrieval combines Amazon’s web index with a structured knowledge graph for verified facts that go beyond crawled web results alone.
The enterprise constraint the service is designed around is data governance: customer queries stay within AWS and are never routed to external search API providers. One early customer noted AWS “keep[s] queries within our trusted AWS environment,” which removes a data residency objection that applies to alternatives that proxy through third-party APIs.
Integration is via Model Context Protocol. AgentCore Web Search exposes a built-in connector on Bedrock AgentCore Gateway, accessible from Python, the MCP Python SDK, the Strands MCP Client, and MCP Inspector for testing. No custom provider code is required for MCP-compatible frameworks.
What This Means for Platform Teams
Enterprise teams currently ground agents via third-party search APIs or build their own web retrieval layer. The third-party approach introduces external data egress; the custom-built approach introduces maintenance overhead. AgentCore Web Search is AWS’s managed alternative to both.
At $7 per 1,000 queries, the cost at 10,000 queries per month is $70 — within infrastructure budget for most active enterprise agent deployments. The pricing is metered and predictable, unlike third-party API arrangements that vary by tier and usage pattern.
The knowledge graph integration is the differentiator worth evaluating. Web index retrieval returns what’s indexed; structured knowledge graph data returns verified facts at lower latency than crawled results. For agents handling product information, company facts, or regulatory data, the knowledge graph component may outperform raw web search for a meaningful subset of queries.
Current availability is US East (N. Virginia) only, with additional regions on the AWS regional capabilities roadmap.
The So What
AgentCore Web Search gives enterprise teams a grounding API that stays inside AWS, simplifies compliance posture for regulated industries, and requires no custom integration for MCP-compatible agent frameworks. The $7/1K pricing is predictable at production query volumes. For teams currently routing agent web queries through external providers or managing a custom retrieval layer, this is worth a benchmarking pass — particularly if data residency or MCP-native integration is already a requirement.
Source: AWS Blog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →