Agentic observability with Amazon OpenSearch Service MCP Apps
aws ·
Amazon OpenSearch Service MCP Apps extend the Model Context Protocol so that AI agent tool calls return both a text summary and an interactive visualization, such as a trace waterfall or service map, rendered directly in the assistant's chat window. This addresses the verification bottleneck where engineers had to leave their IDE, open a separate observability UI, and manually re-run queries to confirm an agent's findings. The architecture uses a local MCP server that bridges the agentic IDE and an OpenSearch UI application, executing authenticated queries against connected data sources like OpenSearch domains, serverless collections, and Amazon Managed Service for Prometheus. The dual response includes a deterministic, server-side rendered visualization, ensuring users see actual query results rather than relying on the AI's interpretation. Available MCP Apps support triage, log investigation, trace investigation, metrics, service performance, topology, and specialized tools for AI/agent observability and stack health. Setup requires an OpenSearch UI Observability workspace, a compatible IDE, Node.js 22 or later, and AWS credentials with es:ESHttpGet and es:ESHttpPost permissions, followed by downloading the MCP server and adding it to the IDE's MCP configuration.
aws