Amazon Web Services published a July 29, 2026 machine-learning blog post describing how Amazon Bedrock AgentCore can generate autonomous business insights by coordinating AI agents with Model Context Protocol (MCP) servers. The example centers on Sarah Chen, who manages 12 assembly lines and 2,000 machines, and needs to know before a 10 AM production review which lines need attention that week. In the scenario, Line 4’s motor temperature runs 12°C above baseline for three days, Machine 42 had a bearing replaced eight months ago under a 12-month warranty, and the operating-hours log shows it has run at 130% rated capacity since January.

The manufacturing example shows how disconnected systems slow decisions. Sarah checks 30-day OEE trends and finds 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, but Line 9 shows no defect signal yet. She emails three supervisors and reaches a confident answer by 11:15 AM. The article contrasts that with Raj Patel, who lacks analytics-dashboard access for Line 7 and works from a PDF export that is three days stale, and Priya Nair, who waits four hours for a CSV about vibration on Machine 42.

AWS frames the problem as enterprises using five to eight operational and analytical systems daily, each with separate logins, query interfaces and access models. Traditional BI dashboards show yesterday’s activity, while single AI assistants answer isolated questions. AWS says custom multi-agent frameworks can span a full technology stack, but require months of engineering for custom connectors, session isolation, memory infrastructure, security code, scaling logic and orchestration logic before the first business question is answered.

Amazon Bedrock AgentCore is presented as a configuration-based approach. AWS says teams connect existing systems with pre-built MCP server connectors, define visibility rules in plain English, and ask natural-language questions while AgentCore handles orchestration, security, memory and scaling. In the proposed architecture, every user interaction passes through Amazon Bedrock, Amazon Bedrock AgentCore authentication, policy, memory and the agent before reaching data sources through MCP connectors.

The architecture has five layers: users, Amazon Bedrock AgentCore, pre-built MCP server connectors, data infrastructure and metadata discovery through a Semantic Layer powered by SageMaker Data Catalog. The agent can route queries to equipment, IoT telemetry, supply chain, analytics or custom low-code connectors, which then pull data from SageMaker Lakehouse, Redshift, S3 Tables, OpenSearch or Aurora. AWS says new sources can be onboarded by registering them in the Data Catalog rather than writing custom integration code, and a complete working implementation is available on GitHub.

The data foundation uses Amazon SageMaker Lakehouse, an open lakehouse architecture compatible with Apache Iceberg that unifies Amazon S3 data lakes, S3 Tables, Amazon Redshift data warehouses and operational data stores connected through zero-ETL integrations. AWS lists pre-built MCP servers for Amazon Redshift, AWS S3 Tables and Amazon Aurora PostgreSQL, plus custom MCP servers for IoT telemetry through Amazon Timestream, quality analytics through Amazon OpenSearch Serverless, and semantic-layer queries through SageMaker Data Catalog metadata.

MCP servers let AI agents discover and call typed tools such as get_equipment_status, detect_anomaly and check_parts_inventory. AWS says AgentCore provides pre-built MCP connectors for Amazon Redshift, Amazon Aurora, Amazon OpenSearch and third-party systems through OAuth, while low-code MCP server templates on Amazon Bedrock AgentCore Runtime cover proprietary or custom systems. Connectivity can include Amazon Aurora, RDS, Amazon MSK, Amazon OpenSearch Service, Amazon S3, Amazon Redshift, Salesforce, SAP and ServiceNow.

AWS also lists trade-offs. MCP servers can run in stateless mode by default or stateful mode for multi-step operations, but stateful servers require session-affinity planning for horizontally scaled deployments. Each added MCP server becomes another endpoint to monitor and version. Amazon Bedrock AgentCore Gateway acts as the single entry point for tool calls, indexes MCP tools after a protocol handshake, routes calls by tool name, and uses organization-scoped and user-scoped caching to reduce latency for repeat queries.

Amazon Web Services published a July 29, 2026 machine-learning blog post describing how Amazon Bedrock AgentCore can generate autonomous business insights by coordinating AI agents with Model Context Protocol (MCP) servers. The example centers on Sarah Chen, who manages 12 assembly lines and 2,000 machines, and needs to know before a 10 AM production review which lines need attention that week. In the scenario, Line 4’s motor temperature runs 12°C above baseline for three days, Machine 42 had a bearing replaced eight months ago under a 12-month warranty, and the operating-hours log shows it has run at 130% rated capacity since January.

The manufacturing example shows how disconnected systems slow decisions. Sarah checks 30-day OEE trends and finds 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, but Line 9 shows no defect signal yet. She emails three supervisors and reaches a confident answer by 11:15 AM. The article contrasts that with Raj Patel, who lacks analytics-dashboard access for Line 7 and works from a PDF export that is three days stale, and Priya Nair, who waits four hours for a CSV about vibration on Machine 42.

AWS frames the problem as enterprises using five to eight operational and analytical systems daily, each with separate logins, query interfaces and access models. Traditional BI dashboards show yesterday’s activity, while single AI assistants answer isolated questions. AWS says custom multi-agent frameworks can span a full technology stack, but require months of engineering for custom connectors, session isolation, memory infrastructure, security code, scaling logic and orchestration logic before the first business question is answered.

Amazon Bedrock AgentCore is presented as a configuration-based approach. AWS says teams connect existing systems with pre-built MCP server connectors, define visibility rules in plain English, and ask natural-language questions while AgentCore handles orchestration, security, memory and scaling. In the proposed architecture, every user interaction passes through Amazon Bedrock, Amazon Bedrock AgentCore authentication, policy, memory and the agent before reaching data sources through MCP connectors.

The architecture has five layers: users, Amazon Bedrock AgentCore, pre-built MCP server connectors, data infrastructure and metadata discovery through a Semantic Layer powered by SageMaker Data Catalog. The agent can route queries to equipment, IoT telemetry, supply chain, analytics or custom low-code connectors, which then pull data from SageMaker Lakehouse, Redshift, S3 Tables, OpenSearch or Aurora. AWS says new sources can be onboarded by registering them in the Data Catalog rather than writing custom integration code, and a complete working implementation is available on GitHub.

The data foundation uses Amazon SageMaker Lakehouse, an open lakehouse architecture compatible with Apache Iceberg that unifies Amazon S3 data lakes, S3 Tables, Amazon Redshift data warehouses and operational data stores connected through zero-ETL integrations. AWS lists pre-built MCP servers for Amazon Redshift, AWS S3 Tables and Amazon Aurora PostgreSQL, plus custom MCP servers for IoT telemetry through Amazon Timestream, quality analytics through Amazon OpenSearch Serverless, and semantic-layer queries through SageMaker Data Catalog metadata.

MCP servers let AI agents discover and call typed tools such as get_equipment_status, detect_anomaly and check_parts_inventory. AWS says AgentCore provides pre-built MCP connectors for Amazon Redshift, Amazon Aurora, Amazon OpenSearch and third-party systems through OAuth, while low-code MCP server templates on Amazon Bedrock AgentCore Runtime cover proprietary or custom systems. Connectivity can include Amazon Aurora, RDS, Amazon MSK, Amazon OpenSearch Service, Amazon S3, Amazon Redshift, Salesforce, SAP and ServiceNow.

AWS also lists trade-offs. MCP servers can run in stateless mode by default or stateful mode for multi-step operations, but stateful servers require session-affinity planning for horizontally scaled deployments. Each added MCP server becomes another endpoint to monitor and version. Amazon Bedrock AgentCore Gateway acts as the single entry point for tool calls, indexes MCP tools after a protocol handshake, routes calls by tool name, and uses organization-scoped and user-scoped caching to reduce latency for repeat queries.