AWS DevOps Agent + LaunchDarkly: Feature Flags as Incident Response
Cloud Platform EngineeringWhen a service starts throwing 5XX errors at 2am, the fastest path to resolution usually isn’t a rollback — it’s a flag flip. The new integration between AWS DevOps Agent and LaunchDarkly’s MCP server makes that path agentic.
The Manual Coordination Problem
Feature flag systems and incident response tools have operated in separate lanes. When something breaks, engineers manually identify which flags are relevant, assess whether reverting a recent flag change would help, and coordinate the change across teams. That process adds minutes of diagnostic latency at the worst possible time.
The AWS DevOps Agent integration closes that gap. The agent connects to LaunchDarkly’s hosted MCP server using standardized tool definitions, enabling it to query current flag state and targeting rules without custom integration code — and to surface actionable containment options directly in the incident workflow.
Two Agentic Use Cases
Pre-deployment defense: Before code ships, DevOps Agent performs release readiness reviews, classifying changes by risk tier (Critical, High, Moderate). When it detects high-risk changes — payment logic, authentication flows, database schema changes — it recommends wrapping them in feature flags and generates the flag configuration, phased rollout strategy, and kill-switch behavior as part of the review output. The recommendation arrives before the incident exists, not after.
Incident response offense: During active incidents, the agent correlates affected services with recent deployments, queries LaunchDarkly for current flag state, and produces structured mitigation plans. The plan format is specific: Prepare (document baselines), Execute (revert the specific flag change), Verify (confirm error rates return to baseline). The output is actionable, not a list of things to investigate.
The Concrete Example
The AWS blog documents a bot-service experiencing sustained 5XX errors. DevOps Agent identifies that the bot-mutation-orchestration-timeout-ms flag had been changed from 2000ms to 30ms, causing ReadTimeout exceptions across the service. The agent recommends reverting the flag to its default variation — restoring sufficient timeout budget without a code deployment. Incident resolved in minutes.
This is the core value of flag-based containment over rollback: no deployment pipeline, no PR merge, no artifact build. The fix is a state change that an agent can both identify and structure into an executable plan.
The So What
Teams running LaunchDarkly alongside AWS infrastructure now have a direct path to agentic flag orchestration without building a custom integration. The MCP protocol is doing real work here: LaunchDarkly hosts its own MCP server, DevOps Agent calls it using standardized tool definitions, and the resulting recommendation layer sits on top of both systems without requiring either to know about the other’s internals. For platform engineering teams evaluating how to reduce mean time to recovery, this integration is worth testing — flag-based containment is already faster than rollback, and agentic orchestration removes the diagnostic step that precedes it.
Source: AWS DevOps Blog
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →