Announcing Web Search on Amazon Bedrock AgentCore: Ground Your AI Agents in Current, Accurate Web Knowledge

Enterprise AI agents require access to current information to execute tasks accurately, yet extending agentic workflows to the open web historically introduces significant operational complexity and data governance risks. The general availability of Web Search on Amazon Bedrock AgentCore provides a fully managed, secure architectural component that bridges agent reasoning directly with live, cited web knowledge while enforcing strict data residency controls.

The feature is exposed as a built-in connector target on the Bedrock AgentCore Gateway, communicating via the Model Context Protocol (MCP). When an agent issues a natural-language query, the Web Search tool retrieves semantically relevant snippets, source URLs, titles, and publication dates from Amazon's multi-source grounding infrastructure. This infrastructure combines Amazon's broad web index with the Amazon Knowledge Graph, which supplies verified factual data to enhance result accuracy over traditional web search alone. A core architectural advantage inherited from Amazon's broader search investments—powering experiences across Alexa+, Amazon Quick, and Kiro—is the zero data egress model. Unlike configurations that forward user prompts or retrieval queries to third-party search API providers external to AWS, Web Search on Bedrock AgentCore executes entirely within the customer's secured AWS environment. This design directly supports enterprise governance policies by ensuring sensitive context and retrieval requests never leave the AWS network boundary.

Implementation follows a straightforward configuration path:

• Provision or select an existing Bedrock AgentCore Gateway.
• Add a new tool target, selecting MCP target as the target protocol and Connectors as the target type.
• Choose the preconfigured Web Search tool target.

Developers can then interact with the Web Search tool programmatically via the AWS API, AWS CLI, or using the MCP Inspector for interactive testing and debugging. This abstraction eliminates the need for platform teams to build, configure, or manage third-party search engine APIs or custom web scraping infrastructure, shifting the operational burden to a fully managed AWS capability.

Enterprise engineering teams in controlled early access validated this architecture. At Benchling, the tool enables grounding in both institutional scientific data and published literature within a governed environment. Gen Digital's Norton Revamp leverages the Web Search tool to generate content ideas grounded in current events, specifically citing the value of queries remaining within the AWS trust boundary. The consumption-based pricing is $7 per 1,000 search queries, with no upfront commitments required.

For enterprise platform teams, Web Search on Amazon Bedrock AgentCore transforms web grounding from a high-touch infrastructure concern into a managed security control, allowing agent development to focus on business logic and reasoning orchestration rather than search pipeline maintenance.

Web Search on Amazon Bedrock AgentCore functions as a fully managed integration that enables AI agents to ground responses in real-time, cited information while maintaining data residency within a customer’s secured AWS environment. The architecture leverages a built-in connector target hosted on the Bedrock AgentCore Gateway, which utilizes the Model Context Protocol (MCP) to facilitate standardized communication between the agentic orchestration layer and external search resources.

The operational workflow follows a structured retrieval sequence:

• Request Processing: The agent initiates a natural-language query sent via the MCP interface to the Bedrock AgentCore Gateway.
• Multi-Source Grounding: The system performs a dual-layer retrieval, combining Amazon’s comprehensive web index with structured, verified facts from the Amazon Knowledge Graph. This synthesis ensures responses are informed by both general web context and high-fidelity, curated data.
• Data Return: The service returns a structured payload containing the most relevant snippets, source URLs, document titles, and publication dates.
• Reasoning: The agent processes these returned metadata points, allowing the model to perform informed inference and generate responses that extend beyond its static training data.

Engineers can implement this by configuring an MCP-compliant target within the Bedrock AgentCore console. Once the Gateway resource URL is established, developers can validate connectivity and query execution using the MCP Inspector, an interactive debugging tool. Interaction is facilitated through various client interfaces, including the MCP Python SDK or direct API requests.

By utilizing the MCP standard, organizations can enforce enterprise governance policies, as search queries and user prompts do not traverse external third-party search APIs. This design allows for the seamless inclusion of current developments in automated workflows while ensuring that all retrieval traffic remains contained within the trusted AWS boundary.

Web Search on Amazon Bedrock AgentCore is implemented as a fully managed, built-in connector target on the Bedrock AgentCore Gateway, utilizing the Model Context Protocol (MCP). This architecture enables agent invocations to be grounded in current web knowledge while maintaining a zero-data-egress posture within the customer's secured AWS environment. The underlying infrastructure combines Amazon's web index with structured Knowledge Graph data, allowing agents to receive cited snippets, source URLs, titles, and publication dates in response to natural-language queries.

To activate the feature, navigate to the Bedrock AgentCore console and create a new Gateway. Configure the Web Search tool target using the following options:

• Select MCP target as the target protocol.
• Select Connectors as the target type.
• Select the Web Search preconfigured target from the available connectors.

After creation, the Web Search tool target is visible on the gateway's detail page and can also be added to an existing gateway. Once the Gateway URL is provisioned, you can interact with the Web Search tool programmatically via API calls, the AWS Command Line Interface (CLI), or the MCP Inspector. Use the View invocation code section in the console to access sample snippets for Python API requests, the MCP Python SDK, or the Strands MCP Client.

The MCP Inspector provides an interactive developer environment for testing and debugging. Connect the Inspector to the Gateway's resource URL; the Web Search tool is listed for each connector target on the gateway. To execute a query, enter a natural-language search string directly into the tool's input field and select Run Tool. The output returns the structured metadata—snippets, source URLs, titles, and publication dates—that the agent reasons over to produce a grounded response.

Pricing follows a usage-based model at $7.00 per 1,000 search queries submitted by agents, with no upfront commitments. This removes the overhead of managing external search APIs and custom index infrastructure, allowing teams to enforce governance policies directly within the AWS account.

The capacity for AI agents to retrieve and act upon current information without compromising enterprise security boundaries represents a significant architectural advance. The Web Search tool on Amazon Bedrock AgentCore addresses this by providing a fully managed mechanism for grounded responses, utilizing a built-in connector target on the Bedrock AgentCore Gateway via the Model Context Protocol (MCP). This ensures that entire query payloads and retrieved context remain within the customer's AWS environment, eliminating data egress to external search APIs.

Rather than relying solely on a static web index, the tool employs a multi-source grounding approach. It combines Amazon's general web index with the Amazon Knowledge Graph, which contains verified facts. This allows an agent to retrieve not just relevant snippets and source URLs, but also structured data, leading to higher accuracy in generated responses without sending internal prompt intent to third-party providers.

Benchling: Secure Hybrid Grounding for Scientific R&D
Early access customer Benchling illustrates the value of this architecture for regulated research environments. Nicholas Larus-Stone, Head of AI Agents at Benchling, noted that scientists can now ground AI responses in both institutional data within the Benchling platform and the latest published literature. He stated that the result is "hypothesis generation done right." The critical technical advantage is that the Web Search tool provides a "secure, governed environment to bring that high quality published data into their workflows without compromising how they manage their data." This directly addresses the requirement for zero data egress when augmenting private datasets with public knowledge.

Gen Digital: Content Grounding Within the Trusted Perimeter
Similarly, Gen Digital leverages the Web Search tool for its Norton Revamp service. Iskander Sanchez-Rola, Senior Director of AI & Innovation at Gen Digital, explained that the tool enables professionals to build online reputation with "current, grounded content ideas shaped by what's actually happening in the world today." He explicitly valued that "AWS uses its own search index and keep[s] queries within our trusted AWS environment." For enterprise security teams, this architectural distinction means that no user prompts or retrieval queries are forwarded to external search API providers outside the AWS boundary.

Enterprise Governance Considerations
The governance advantages stem directly from the zero data egress design. Enterprise architects can implement AI agents that access current web knowledge without:

• Exposing internal query intent or user context to third-party search engines.
• Managing custom web scraping or dedicated search infrastructure.
• Configuring complex network egress rules or VPC endpoints for agent tooling.

By integrating the MCP standard and relying on Amazon's own search infrastructure, organizations can meet enterprise governance policies effectively. Agents reason over current events or niche domains beyond the model's training cutoff while maintaining full operational visibility and control over data flow. This represents a shift toward secure, hybrid grounding where internal institutional knowledge and public web data are synthesized within a unified security boundary.

Web Search on Amazon Bedrock AgentCore is generally available in the US East (N. Virginia) Region (us-east-1). There are no upfront commitments or minimum usage requirements, and you can begin using the feature immediately from the Bedrock AgentCore console or API. For roadmap information and future regional expansion, consult the AWS Capabilities by Region page.

Pricing follows a simple, usage-based model: $7 per 1,000 search queries. Each query submitted by an agent to the Web Search tool counts toward this metric, regardless of the number of results returned. There are no tiered brackets, residual fees, or mandatory support contracts.

• Unit of consumption: 1 search query = 1 API call from agent to Web Search tool target.
• Billing granularity: per-query increments; no hourly or monthly minimums.
• Free Tier: new AWS customers receive up to $200 in Free Tier credits, which can offset initial usage.

Practical example: An agent that issues 25,000 search queries in a month would incur 25,000 ÷ 1,000 × $7 = $175. If the account is within its first 12 months and has not exhausted its Free Tier credits, that $175 would be applied against the $200 credit allowance, resulting in no charge for that period. After credits are consumed, the $7 per 1,000 rate applies automatically with no need to re-provision or switch pricing plans.

For complete pricing details—including data transfer costs, API request surcharges (if any), and regional price variations—refer to the Amazon Bedrock AgentCore pricing page. The pricing page also documents any applicable service-level agreements and terms for the Free Tier offer.

This consumption-based approach aligns with standard AWS pricing philosophy: you pay only for what you use, with no over-provisioning or reserved capacity required. For enterprise governance, all queries remain within the customer’s AWS environment and never egress to external search providers, ensuring that billing reflects only the search volume, not data transfer or third-party API fees.

Web Search on Amazon Bedrock AgentCore provides a fully managed mechanism for grounding agent responses in up-to-date, attributed web content without routing user prompts or retrieval queries to external search API providers. The tool operates through a built-in connector target on the Bedrock AgentCore Gateway using the Model Context Protocol (MCP). When an agent issues a natural-language query, the Web Search tool returns relevant snippets, source URLs, titles, and publication dates that the model can reason over to produce a grounded answer. The underlying infrastructure combines Amazon's web index with structured data from the Amazon Knowledge Graph, enabling multi-source retrieval that extends beyond traditional web search alone.

To integrate this capability into your agent workflow:

• Create a Bedrock AgentCore Gateway in the console and select MCP target as the target protocol with Connectors as the target type.
• Choose the Web Search preconfigured tool to retrieve snippets, links, and metadata.
• Interact with the tool via API calls, AWS CLI, the MCP Python SDK, the Strands MCP Client, or the MCP Inspector—a developer utility for testing and debugging MCP servers.

After creating the Gateway, the Web Search tool target appears on the detail page. You can also add it to an existing gateway. For example, when connected through the Gateway resource URL in MCP Inspector, you can enter a query, select Run Tool, and inspect the returned results directly.

Pricing is usage-based at $7 per 1,000 search queries, with no upfront commitments. New AWS customers receive up to $200 in Free Tier credits. Production deployments should plan query volume against this cost model and implement appropriate caching or throttling where necessary.

To begin, visit the Amazon Bedrock AgentCore console and create your first Gateway with the Web Search tool. For feedback, use AWS re:Post for Amazon Bedrock AgentCore or your usual AWS Support contacts. The Bedrock AgentCore Gateway documentation provides API references, invocation code samples, and configuration guidance. Build agents that ground their reasoning in current web knowledge while keeping all retrieval within your secured AWS environment.