Jugael is an MCP server that turns the AI agent you already use into an incident responder. It reads your logs, metrics and recent changes, finds the probable root cause with evidence, and drafts the fix and the postmortem. You stay in control of every action.
> Checkout API is returning 502s. What happened? ⏺ jugael.triage_alerts (service="checkout-api", window="30m") ● P1 1,312 alerts grouped into 3 incidents (98% noise removed) ⏺ jugael.investigate (incident="INC-0142") ├─ logs upstream timeout: payments-svc:8443 ×4,870 ├─ k8s payments-svc OOMKilled, 6 restarts ├─ changes deploy a91f3c · 14:02 · memory limit 512Mi → 256Mi └─ metrics p95 latency 180ms → 9.4s at 14:04 ✓ Probable root cause (confidence: high) Commit a91f3c halved the memory limit of payments-svc. ⚑ Proposed fix — requires your approval 1. Roll back deploy a91f3c 2. Draft postmortem Approve? [y/N]
No new dashboard to learn. Jugael gives your agent the tools and the method; specialized sub-agents do the digging in parallel.
Add Jugael to Claude, Cursor or your own agent and connect your log, metric, Kubernetes and deploy sources with read-only credentials.
Sub-agents for logs, metrics, infrastructure and recent changes test hypotheses at the same time and correlate what they find.
You get the probable root cause with cited evidence, a proposed remediation for you to approve, and a postmortem draft.
Start with incident response. Use the same analyzer to prevent the next outage, cut costs and prove compliance.
Group, deduplicate and rank hundreds of alerts so on-call sees the few that matter.
Correlate logs, metrics, traces and changes into a probable cause with evidence.
Rollback, restart or scale proposals, executed only after human approval.
Timeline, impact, contributing factors and action items, ready to export.
Every resolved incident becomes a reusable runbook your team and agents can follow.
Flag a deploy that looks like one that caused an incident before.
Explain OOMKilled, CrashLoopBackOff, failed probes and noisy cluster events.
Read the failed job, find why it broke and suggest the fix as a pull request.
Plain-language summaries of what happened overnight, for engineers and leaders.
Spot over-provisioned services by joining usage metrics with scaling history.
Every query and action is logged, useful evidence for SOC 2 and ISO 27001.
Detect suspicious patterns like auth failures and hand them to your security team.
Jugael speaks the open Model Context Protocol, so any MCP-compatible client can use it. Paste the config, sign in with OAuth, and ask your agent about your last incident.
{
"mcpServers": {
"jugael": {
"type": "http",
"url": "https://mcp.jugael.de/mcp"
}
}
}
Endpoint available to beta users. Config shown is illustrative and may change before general availability.
Connectors are rolling out during the beta. Tell us your stack and we will prioritize it.
Autonomy grows one runbook at a time, only when you decide it has earned it.
Read-only investigation. Jugael explains and proposes; humans act.
Jugael runs a specific action after a person approves it in the chat or incident channel.
Only for runbooks you promote, with limits, rollback and a full audit trail.
Incident analysis is a reasoning problem over huge, messy inputs. Jugael uses the Claude API to orchestrate its sub-agents and reach a conclusion you can verify.
Get senior-SRE-style investigations without hiring a dedicated reliability team.
Investigate incidents across many client environments from a single agent.
Debug production from Claude Code or Cursor, right next to the code that broke.
The Model Context Protocol is an open standard that lets AI assistants use external tools. Jugael is an MCP server: once you add it, your agent gains incident-analysis tools it can call on its own.
No. Jugael reads from the observability, Kubernetes and deploy tools you already use. It adds an investigation layer on top.
Not by default. It starts read-only. Write actions are opt-in, scoped, and require human approval unless you explicitly promote a runbook to auto-remediation.
Only the sources you connect. Sensitive values are masked before analysis, every access is logged, and teams that need it can self-host the server on their own infrastructure.
Any MCP-compatible client, including Claude, Claude Code, Cursor, VS Code and Windsurf, plus custom agents built with an MCP SDK.
The private beta is free. We will share pricing with beta users before general availability.
We are onboarding a small group of teams. Tell us about your stack and we will reach out.
Or write to us at hello@jugael.de