SRE TriageCore

Inspiration

I was inspired by the disorientation on-call engineers face during sudden 3:00 AM alerts, where hunting across dashboards and diffs under pressure leads to slow recovery and dangerous operational mistakes.

What it does

It synthesizes real-time Prometheus telemetry spikes, correlates breaking GitHub pull request diffs, and extracts targeted Markdown runbook steps into a concise voice briefing while denying mutating actions like unsafe container rollbacks.

How we built it

I built it using the Model Context Protocol (MCP) over Streamable Server-Sent Events, an Express TypeScript server, Prometheus PromQL querying, the Octokit GitHub API, and an Alexa+ developer audio harness with the Web Speech API.

Challenges we ran into

Enforcing the strict RFC 9728 Protected Resource Metadata read-only boundary across all streaming tool dispatchers without breaking standard diagnostics was our toughest architectural hurdle.

Accomplishments that we're proud of

I am proud of creating a sub-second ambient triage workflow that answers an alert verbally and displays clean, correlated unified code diffs without touching a single button.

What we learned

I learned how Model Context Protocol tool routing dramatically simplifies agentic development when coupled with deterministic, RFC-compliant security metadata guards.

What's next for SRE TriageCore

Next, I plan to support multi-service topological dependency tracing and turn the web harness into a native Alexa voice skill with AWS CloudWatch metric hooks.

Built With

• docker
• express.js
• mcp
• prometheus
• typescript