Amazon OpenSearch Service launches MCP Apps for agentic observability
Investigate incidents with AI agents inside VS Code or Claude Desktop, with interactive OpenSearch visualizations rendered inline—no browser switching needed.
View original announcement →Visual Summary
graph TD
A{{OpenSearch MCP Apps}}:::announced
B((Developer IDE)):::external
C(Amazon OpenSearch Service):::storage
D(Amazon Managed Prometheus):::storage
E([Dual Response Engine]):::feature
F([Interactive Visualizations]):::feature
G([Cross-Signal Correlation]):::feature
H(OpenSearch UI):::compute
B ==>|"MCP tool calls"| A
A -->|"queries logs/traces"| C
A -->|"queries metrics"| D
A -->|"returns"| E
E -->|"text summary"| B
E -->|"renders"| F
A -.->|"correlates signals"| G
A -->|"built on"| H
H -->|"connects to"| C
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
Amazon OpenSearch Service now supports MCP Apps, enabling AI agents running in local agentic IDEs such as Claude Desktop and VS Code to perform observability workflows—including incident investigation, root cause analysis, and distributed trace exploration—directly against OpenSearch domains, serverless collections, and Amazon Managed Service for Prometheus. Each MCP tool call returns a dual response: a concise text summary for the agent to reason over and an interactive visualization rendered inline in the conversation thread. This brings the full observability signal set (logs, traces, metrics, and alerts) into a single, agent-driven conversational workflow without requiring engineers to switch contexts.
How It Works
- MCP Server Setup: Users configure an MCP server that connects their local agentic IDE (Claude Desktop, VS Code, or any MCP-compatible client) to their OpenSearch domains, serverless collections, and Amazon Managed Service for Prometheus endpoints via an
.mcpbconfiguration file. - Dual-Response Tool Calls: Every MCP App tool invocation returns two artifacts simultaneously—a structured text summary the AI agent uses for reasoning and chaining next steps, and an interactive visualization rendered directly in the conversation thread for human review.
- Available Tool Categories: The MCP App toolset covers log investigation, metrics analysis, trace investigation, service performance, topology/service maps, dynamic visualizations, agent health, cluster health, and instrumentation scoring.
- Signal Correlation: The agent can correlate across multiple observability signals (logs, traces, metrics, alerts) within a single conversation, enabling end-to-end incident workflows from alert triage through root cause analysis.
- Visualization Types: Interactive outputs include firing alert timelines with severity breakdowns, distributed trace span hierarchies, service dependency call graphs (with edge width for call volume and color for error rate), and PromQL metric charts (line, bar, area, or metric formats).
- Underlying Platform: The experience is built on OpenSearch UI, a cloud-hosted, multi-source operational analytics layer that supports OpenSearch domains, serverless collections, CloudWatch, Security Lake, and S3 as data sources.
Why It's Important
- Eliminates Context Switching: Engineers can investigate production incidents entirely within their local IDE, removing the need to jump between dashboards, terminal windows, and separate observability consoles during high-pressure incidents.
- Accelerates Root Cause Analysis: By combining AI-driven reasoning with interactive visualizations in a single thread, teams can move from alert firing to root cause identification faster than traditional manual workflows.
- Human-in-the-Loop Verification: The dual-response model ensures that AI-generated conclusions are always paired with verifiable visual evidence, reducing the risk of acting on incorrect agent inferences.
- Unified Observability Signal Coverage: Covering logs, metrics, traces, and alerts in one conversational workflow addresses the fragmentation problem that has historically made cross-signal correlation slow and error-prone.
- Lowers the Observability Skill Floor: Natural language–driven investigation via an AI agent makes deep observability analysis more accessible to developers who are not expert in query languages like PPL, PromQL, or Lucene.
- Extensible to Managed Prometheus: Native integration with Amazon Managed Service for Prometheus means teams using Prometheus-based metrics stacks get the same agentic experience without additional instrumentation changes.
How It's Different
- Versus Traditional OpenSearch Dashboards: OpenSearch Dashboards requires users to manually build and navigate dashboards within a browser; MCP Apps bring the data to the developer's existing IDE via conversational AI, with no manual dashboard construction required.
- Versus OpenSearch UI Alone: OpenSearch UI is a cloud-hosted web interface; MCP Apps extend that observability capability into local agentic environments, enabling programmatic, agent-driven investigation rather than point-and-click exploration.
- Versus Generic LLM + Observability Integrations: Rather than sending raw data to an LLM and receiving only text, MCP Apps produce structured interactive visualizations alongside text, giving engineers a verifiable artifact rather than just an AI narrative.
- Versus Standalone Prometheus/Grafana Workflows: PromQL metric charts, service maps, and trace views are all surfaced within the same conversation thread, eliminating the need to context-switch to Grafana or Jaeger for different signal types.
- Dual-Response Architecture: The explicit separation of "text for agent reasoning" and "visualization for human review" is a deliberate design choice that most observability integrations do not implement, making the human-AI collaboration more structured and auditable.
When to Prefer It
- Active Incident Response: When an on-call engineer needs to triage a firing alert, correlate it with traces and logs, and identify root cause quickly without leaving their IDE.
- Distributed System Debugging: When investigating latency or error spikes in microservices architectures where understanding service topology and trace span hierarchies is essential.
- AI-Augmented SRE Workflows: When teams want to use AI agents to automate the initial investigation steps (alert review, log pattern detection, metric anomaly identification) while retaining human oversight via inline visualizations.
- Multi-Signal Correlation: When an incident requires correlating Prometheus metrics, OpenSearch logs, and distributed traces simultaneously—a workflow that is cumbersome across separate tools.
- Developer-Led Observability: When application developers (not dedicated SREs) need to investigate production issues from their primary development environment (VS Code) without learning a separate observability platform.
- Instrumentation Auditing: When teams want to assess the completeness and quality of their observability instrumentation using the instrumentation scoring tool before or after a deployment.
Availability
- GA Status: Generally available as of June 10, 2026.
- Regional Availability: Available in all AWS Regions where Amazon OpenSearch UI is offered; no specific region exclusions were called out in the announcement.
- Supported IDEs: Compatible with any MCP-compatible agentic IDE, with Claude Desktop and VS Code explicitly cited as supported clients.
- Data Source Support: Works with Amazon OpenSearch Service domains, OpenSearch Serverless collections, and Amazon Managed Service for Prometheus.
- Pricing: No separate pricing was announced for MCP Apps; costs are expected to follow standard Amazon OpenSearch Service and Amazon Managed Service for Prometheus pricing for the underlying data queries and storage.
- Prerequisites: Requires an existing OpenSearch domain or serverless collection, a configured MCP server, and a compatible agentic IDE; setup is documented in the OpenSearch Agentic Observability with MCP Apps developer guide.
