Announcing Web Search on Amazon Bedrock AgentCore for Agentic Web Retrieval
AWS agents can now search the live web natively inside your AWS environment — no third-party APIs, no data egress, no extra setup.
View original announcement →Visual Summary
graph TD
A{{Web Search on AgentCore}}:::announced
B((AI Agent)):::external
C(AgentCore Gateway):::compute
D([MCP Protocol]):::feature
E(Amazon Web Index):::storage
F(Amazon Knowledge Graph):::storage
G([Zero Data Egress]):::feature
H([Ranked Excerpts]):::feature
I(Amazon Bedrock):::compute
B ==>|"natural-language query"| C
C ==>|"routes via"| D
D ==>|"invokes"| A
A -->|"crawls"| E
A -->|"retrieves facts"| F
A -->|"returns"| H
H -->|"reasons over"| I
I -->|"grounded response"| B
A -.->|"enforces"| G
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
AWS has announced the general availability of Web Search on Amazon Bedrock AgentCore, a fully managed tool that enables AI agents to retrieve current, accurate web information without leaving the AWS environment. Built on Amazon's own search infrastructure and exposed via the Model Context Protocol (MCP) through AgentCore Gateway, it combines a live web index with structured knowledge graph data to ground agent responses in up-to-date facts. This eliminates the need for developers to integrate and manage third-party search providers when building agentic applications on Bedrock AgentCore.
How It Works
- MCP-based connector: Web Search is exposed as a built-in connector target on the AgentCore Gateway using the Model Context Protocol (MCP), so agents interact with it through a standardized, natural-language query interface.
- Multi-source grounding: Results are drawn from two complementary sources — Amazon's own web index (a crawled, ranked web corpus) and a structured Amazon Knowledge Graph containing entity data and verified facts.
- Agentic-optimized output: The tool returns ranked, high-value excerpts (snippets), source URLs, titles, and publication dates, designed to maximize relevant information per token for downstream model reasoning.
- Zero data egress: All query processing and result retrieval remain within the customer's secured AWS environment; no user prompts or retrieval queries are sent to external third-party search APIs.
- Gateway integration: Developers configure a Web Search tool target in the AgentCore Gateway console (choosing MCP target protocol and Connectors as the target type), after which the endpoint is accessible via Python SDK, AWS SDK (boto3), CLI, or MCP Inspector.
- Invocation flow: The agent sends a natural-language query → AgentCore Gateway routes it to the Web Search connector → results are returned as structured JSON with snippets, URLs, titles, and dates → the model reasons over results to produce a grounded response.
Why It's Important
- Eliminates undifferentiated heavy lifting: Previously, developers had to integrate external search APIs (e.g., Bing, Google), build custom orchestration, manage separate authentication, billing, and compliance contracts — all of which is now handled by AWS.
- Keeps agents current: LLM training data has a knowledge cutoff; Web Search allows agents to answer questions about recent events, current prices, latest documentation, or breaking news without retraining or fine-tuning.
- Enterprise compliance by default: Because data never leaves AWS, organizations subject to strict data residency, sovereignty, or governance policies can use web-grounded agents without violating those policies.
- Reduces hallucination risk: By grounding responses in cited, real-time web sources and verified knowledge graph facts, agents are less likely to fabricate answers on topics outside their training data.
- Accelerates agent development: Developers can add real-time web grounding to an agent in minutes via console or SDK, rather than spending weeks on search infrastructure integration.
- Battle-tested infrastructure: The underlying search stack powers Alexa+, Amazon Q Business, and Kiro, providing production-grade reliability and relevance tuning from day one.
How It's Different
- Native AWS integration vs. third-party APIs: Unlike Bing Search API, Google Custom Search, or Brave Search, Web Search on AgentCore requires no external accounts, API keys, or cross-provider billing — it is provisioned and billed entirely within AWS.
- Zero data egress guarantee: Competing approaches that call external search providers send user queries outside AWS; AgentCore Web Search keeps all data within the customer's AWS boundary, a meaningful differentiator for regulated industries.
- Knowledge graph augmentation: Standard web search tools return only crawled web pages; AgentCore Web Search additionally queries a structured knowledge graph for entity-level, verified facts, improving accuracy on factual queries.
- MCP-native design: The tool is purpose-built for the Model Context Protocol, making it a first-class citizen in MCP-based agentic architectures rather than a wrapper around a REST API.
- Agentic retrieval optimization: Results are pre-ranked and excerpt-optimized for token efficiency in agent reasoning loops, rather than returning full web pages or raw HTML that agents must parse.
- Unified AgentCore ecosystem: Web Search integrates directly with AgentCore Memory, Runtime, Identity, and Browser components, enabling richer multi-tool agent workflows without additional glue code.
When to Prefer It
- Real-time information needs: Use Web Search when your agent must answer questions about current events, recent product releases, live pricing, or any topic that postdates the model's training cutoff.
- Enterprise compliance requirements: Prefer this tool when your organization prohibits sending user data or query content to third-party vendors outside AWS, such as in financial services, healthcare, or government workloads.
- Factual entity queries: Choose Web Search when agents need precise, verified facts about people, organizations, products, or locations — where the knowledge graph augmentation provides higher accuracy than raw web results.
- Rapid prototyping on AgentCore: Use it when you want to add web grounding to an existing AgentCore-based agent quickly, without architecting a separate search integration layer.
- Cost-consolidated AWS environments: Prefer this when your team wants a single AWS bill and unified IAM/security model across all agent tooling, rather than managing separate vendor relationships.
- Citation-required responses: Use Web Search when your agent must return sourced, attributable answers with URLs and publication dates, such as in research assistants, compliance tools, or customer-facing Q&A bots.
- High-volume agentic workloads: Prefer it when you need a managed, scalable search backend that handles traffic spikes without requiring you to manage rate limits or quotas with an external provider.
Availability
- GA status: Generally available as of June 16, 2026 — this is a production-ready release, not a preview.
- Supported regions: Currently available in a single AWS Region: US East (N. Virginia) (us-east-1); additional regions have not been announced.
- Access method: Accessible via the Amazon Bedrock AgentCore console, AgentCore Python SDK, AWS SDK for Python (boto3), CLI, and MCP Inspector.
- Pricing: Specific pricing details were not disclosed in the announcement; consult the AgentCore pricing page or AWS documentation for current rates.
- Prerequisites: Requires an AgentCore Gateway to be configured; Web Search is added as a connector target (MCP protocol, Connectors target type) within that gateway.
- Limitations: Single-region availability at launch may be a constraint for workloads requiring multi-region or non-US data residency; no mention of support for non-English queries or international web indexes at this time.
