Amazon Bedrock AgentCore Announces Web Search Feature: Combining Up-to-Date Web Knowledge with AI Agents

Amazon has officially launched the Web Search feature in Amazon Bedrock AgentCore, helping AI agents generate answers based on the latest web knowledge. This feature is a fully managed tool that can fetch up-to-date information within a customer's secured AWS environment without data leakage.

Web Search uses the Model Context Protocol (MCP) to leverage the built-in connector of Bedrock AgentCore Gateway. When an agent sends a query in natural language, it returns relevant snippets, source URLs, titles, publication dates, and more, supporting the model in generating grounded answers.

Key features are as follows:

• Multi-source grounding: Combines Amazon's web index with structured Knowledge Graph data to provide more accurate information than general web search.
• Security and governance: Enables compliance with enterprise governance policies without sending user prompts to external search API providers outside of AWS.
• Developer convenience: The feature can be easily added through MCP target and Connectors configuration in the Bedrock AgentCore console without infrastructure management.

Developers can test and debug the Web Search tool through the Python SDK, CLI, or MCP Inspector, enabling agents to take appropriate action even for recent events outside the scope of the model's training data.