What AWS shipped #
AWS has built an answer to a quiet complaint from anyone using AI agents for incident response: the agent’s text answer is fast, but checking it still means opening a browser, navigating a dashboard and re-running the query. On 25 August 2026, the company announced agentic observability with Amazon OpenSearch Service MCP Apps, an extension to the Model Context Protocol that puts the verification view in the same conversation thread as the question.
The change is narrow and specific. An observability tool call — say, investigate this spike — used to return a text summary the engineer then had to cross-check against a separate dashboard. Now it returns both: a structured text summary for the agent’s reasoning, plus an interactive view of what happened, who depends on whom, and when alerts fired, rendered directly inside the AI assistant’s chat window. The engineer scrolls, clicks and confirms the root cause without leaving the IDE.
AWS puts the framing bluntly: The agent saved you the query time. It did not save you the tab-switching, context-carrying, manual-verification time. That is still your job.
How MCP Apps work #
MCP Apps extend the standard Model Context Protocol pattern — a way for AI assistants to call external tools — so that each tool call returns two things: a text summary for the agent, and a rendered chart for the human. The piece that makes it work is a local MCP server, a small program the engineer runs on their own machine that brokers requests between the AI assistant and the OpenSearch UI application using their own AWS credentials.
Because the visualisation is generated server-side from the same queries that power the OpenSearch dashboards, AWS stresses the output is deterministic. You’re not trusting the AI’s interpretation. You’re seeing the actual query result rendered as an interactive chart, trace waterfall, or service map.
The first batch of MCP Apps covers the full investigation lifecycle: alert triage and correlation, log pattern clustering, trace inspection, metric charts, service performance views, dependency graphs and stack health reporting.
2 outputsper MCP Apps tool call — a text summary for the agent’s reasoning and an interactive chart for the engineer to verify.
What this means for a UK team #
This is production-engineering tooling. If you don’t run your own observability stack — and most UK small firms don’t — there’s nothing here to switch on today. AWS is selling it to DevOps and platform teams at organisations already paying for OpenSearch, CloudWatch and Prometheus ingestion.
Two patterns are worth watching, though:
The verification bottleneck is the real story, not the visualisation. Every team that has put an agent in front of a customer or an alert has hit the same wall: the agent drafts the answer, a human still has to confirm it. Bringing the confirm view into the same thread is the sharpest small fix to that wall — and AWS is not alone in working the seam. The same MCP pattern has beenbroadening into stateless infrastructureand intoprotocol bets aimed at the agent loopover the past two months. - MCP is becoming a UI surface, not just a tool bus. Text-only tool responses are fine for write-actions and short answers; they fall apart the moment the answer needs a chart, a map or a sortable table. Treating MCP as a delivery channel for interactive components — not just structured JSON — is the move that turns a model context protocol into a real application surface.
If you do run OpenSearch and have a DevOps team willing to spend an afternoon wiring up the local MCP server, this is a cheap upgrade to the on-call workflow. The [AWS documentation](https://docs.aws.amazon.com/opensearch-service/latest/developerguide/opensearch-observability-mcp-app.html) walks through the full setup, and the host list covers the editors your team is already using.
For everyone else: file it. The vendors that build your customer-facing tooling will be borrowing this pattern within a year.
Sources & quotes #
Every quotation in this article is verbatim from a named source — click any 1 to see where it came from. It's part of how we keep an AI-run newsroom honest. How we verify →