Amazon Web Services added domain filtering and published-date filtering to Web Search in Amazon Bedrock AgentCore on August 20, 2026, and expanded the tool to the EU (Ireland) and Asia Pacific (Tokyo) regions. Web Search lets an enterprise AI agent retrieve live, cited web knowledge without customer data leaving the customer’s AWS environment, and AWS made the tool generally available in June 2026 starting in US East (N. Virginia). The tool queries a purpose-built Amazon web index spanning tens of billions of documents refreshed within minutes, connects to agents through the Model Context Protocol (MCP) on Bedrock AgentCore Gateway, and costs $7 per 1,000 queries.
What Domain and Date Filtering Control
Domain filtering restricts an agent’s web searches to a named set of websites, and published-date filtering restricts results to a chosen time window. Both controls operate at the tool level, so the restriction applies to every query the agent issues rather than depending on prompt instructions the model can ignore.
The two filters solve distinct failure modes in agentic retrieval:
- Domain filtering addresses source trust — a compliance team can limit a support agent to regulator sites, vendor documentation, and the company’s own domains, such as an insurer restricting an agent to its own policy pages plus national regulator publications.
- Published-date filtering addresses staleness — a pricing or policy agent can exclude documents published before a cutoff date, so the agent stops citing superseded rates, discontinued products, and withdrawn guidance.
Why the EU and Asia Pacific Expansion Matters
The regional expansion makes Web Search usable by companies whose data-residency rules forbid routing requests through US infrastructure. Before August 20, Web Search ran only in US East (N. Virginia), which excluded workloads governed by EU data-residency commitments and by Japanese in-country processing requirements. EU (Ireland) and Asia Pacific (Tokyo) now run the same tool.
The no-egress design compounds that benefit. Search requests and results stay inside the customer’s AWS account rather than transiting a third-party search vendor, which removes one data-processing agreement from every agent deployment. For businesses assembling agent stacks, this is the difference between a procurement review and a security exception — the same evaluation axis that separates the platforms in our guide to the 12 Best AI Agents for Business Tasks, where data handling decides which vendors reach a shortlist.
How Web Search Compares to Third-Party Search APIs
Web Search competes with the search APIs enterprises currently bolt onto agents, such as Tavily, Brave Search API, and Bing Web Search. AWS’s differentiators are the no-egress guarantee and native MCP integration with Bedrock model deployments, so an agent already running on Bedrock adds live search as a Gateway connector instead of as an external HTTP integration with its own credentials, billing, and audit surface.
No independent latency or result-quality benchmarks compare Web Search against those alternatives, so buyers should test retrieval quality on their own query mix before switching. Pricing did not change in the August update: the $7 per 1,000 queries rate set at launch still applies, which puts 10,000 agent searches per month at $70.
Our take: AWS is closing the gap between “an agent that can search” and “an agent legal will approve.” Data stays in the account, the search surface is bounded by domain, and the results are bounded by date — three of the four questions a risk reviewer asks. The era of stitching a third-party search key into every agent is getting shorter.
For Context
- MCP Goes Stateless: What the New Roadmap Means for AI Agents — the protocol Bedrock AgentCore Gateway uses to expose tools such as Web Search.
- MCP 2026 Spec Is Live: What the Stateless Rewrite Means for AI Tools — the July 28 specification release behind the current connector model.
- Cloudflare Kitesurf: The AI Agent Browser That Cuts Costs 3–7× — a competing approach to giving agents live web access.
Related Reading
- Humans Miss 1 in 3 AI Agent Threats — 40,000-Run Study — why bounding an agent’s tool surface reduces reviewer workload.
- AI Agents Hit Death Valley: 99% Plan Them, Only 10% Ship — the integration and approval friction these managed tools target.
- 15 AI Agent Examples Across Industries — the workloads that need grounded, current answers.
Sources: Web Search in Amazon Bedrock AgentCore adds domain and published-date filtering, AWS What’s New, August 20, 2026; Web Search on Amazon Bedrock AgentCore, AWS What’s New, June 2026.
