AWS Bedrock AgentCore lets businesses query all data systems in plain English

Sarah Chen manages 12 assembly lines and 2,000 machines at a manufacturing facility. Before her 10 AM production review, she needs to answer a straightforward question: which lines need attention this week? Today, that answer takes more than an hour. She starts in the IoT dashboard, where she sees that Line 4's motor temperature is running 12°C above baseline and has been for three days. But the dashboard shows signals, not causes. She pivots to the ERP system to check maintenance history and learns that Machine 42 had a bearing replaced eight months ago, with a 12-month warranty. An operating hours log in a separate database reveals the bearing has been running at 130% rated capacity since January — context that does not appear in the IoT dashboard. She then pulls 30-day OEE (overall equipment effectiveness) trends and finds that Line 4's availability dropped from 94% to 87%, while Line 9's throughput dipped 6%. Line 4's scrap rate jumped 2.3% last Tuesday. She wonders if the problems are related through a shared coolant loop, but that relationship is not modeled in any system. She emails three supervisors and does not get a confident answer until 11:15 AM.

Meanwhile, Line Supervisor Raj Patel is trying to do the same analysis for Line 7 but cannot access the analytics dashboard — he is working from a PDF export three days old. Maintenance Technician Priya Nair flagged a vibration on Machine 42 last week and wants to know if it has worsened, but she does not have credentials to access the data. She asks Raj, who asks Sarah's analyst, and a CSV arrives four hours later. Three people, three access levels, five disconnected systems, the same underlying data. The enterprise is data-rich but insight-poor.

AWS Bedrock AgentCore is designed to eliminate this friction. Instead of building an autonomous intelligence system from scratch — a process requiring months of custom connectors, session isolation, memory infrastructure, security code, scaling logic, and orchestration — users configure one. The architecture has five layers. At the top are users who ask natural language questions. Amazon Bedrock handles intent classification and response synthesis, then routes to Amazon Bedrock AgentCore, which orchestrates execution. The Agent sits at the center, managing authentication, policy enforcement, persistent memory, and access to a governed tool registry. Before selecting which data connector to call, the Agent consults the Semantic Layer, powered by SageMaker Data Catalog, which helps the Agent understand what data exists across the enterprise and where it resides without hard-coded logic. Once the relevant data sources are identified, the Agent routes the query to the appropriate MCP server connector — pre-built for systems like Redshift, Aurora, or OpenSearch, or custom-built using low-code templates for specialized services like Amazon Timestream (for IoT telemetry) or Amazon OpenSearch Serverless (for quality analytics).

The Model Context Protocol (MCP) is an open standard that defines how AI agents discover and invoke tools. In this architecture, each domain of the business — equipment status, IoT telemetry, supply chain, quality management — is represented by a dedicated MCP server exposing typed tools such as get_equipment_status, detect_anomaly, or check_parts_inventory. Pre-built MCP servers are available for Amazon Redshift (warehouse queries), AWS S3 Tables (Iceberg-format historical data), Amazon Aurora PostgreSQL (operational data), and third-party SaaS via OAuth. For proprietary or custom systems, low-code MCP server templates on Amazon Bedrock AgentCore Runtime reduce the effort from full development to configuration. Building a custom MCP server for a service like Timestream requires only defining the tool interface and delegating to the AWS SDK — a few lines of Python using the FastMCP framework to declare a tool, validate input, query the service, and return JSON. Once registered, the custom server behaves identically to a pre-built one; the Agent discovers its tools through the MCP protocol and calls them the same way.

The complete working implementation is available on GitHub. The setup process follows three steps: connect existing systems using pre-built MCP server connectors with no custom integration code, define who can see what using plain-English policy rules that Amazon Bedrock AgentCore translates into enforcement logic, and ask questions in natural language while Amazon Bedrock AgentCore orchestrates the rest. With this model, Sarah, Raj, and Priya can each ask natural language questions and receive synthesized, personalized answers from across the entire technology stack without knowing which system provided which piece of the answer.

The problem Bedrock AgentCore addresses is structural: modern enterprises have distributed data across operational systems, data warehouses, IoT platforms, and analytics tools, but no unified query layer. Today, a manager who needs cross-system context — e.g., is a production line's temperature spike related to a cooling loop failure or a bearing at the end of its warranty? — must manually navigate multiple dashboards, pull data from different teams, and wait hours for answers. The engineering cost of building a custom multi-agent system from scratch is prohibitive: custom connectors, session isolation, memory infrastructure, security, scaling, and orchestration logic. Bedrock AgentCore inverts this model by providing those layers as managed infrastructure, then exposing a single natural language interface.

The architecture is deliberate in its layering: users query at the top, Bedrock AgentCore orchestrates at the center, MCP servers abstract data sources in the middle, and underlying infrastructure (SageMaker Lakehouse, Redshift, S3, Aurora, OpenSearch, Timestream) sits at the foundation. This separation means new data sources can be on-boarded by registering them in the Data Catalog and connecting a pre-built or low-code MCP server, without modifying the agent or orchestration logic. The reliance on the open Model Context Protocol (MCP) standard as the interface between agent and data tools is significant: it avoids vendor lock-in and allows MCP-compatible servers built by third parties to work alongside AWS connectors.

No comments yet. Be the first to share your thoughts!