Home

mgtt

Architecture diagrams go stale, and at 3am the person who drew one is asleep. mgtt turns that diagram into a YAML model and runs one engine over it at every stage of a system's life.

graph LR
  M[model] --> S[simulate in CI]
  S --> D[diagnose at 3am<br/>you or an AI agent]
  D --> R[retrospective:<br/>incident becomes a scenario]
  R --> S

Model¶

You list components, what each depends on, and what healthy means. Types come from providers (Kubernetes, AWS, Docker, Terraform, Tempo, Quickwit), so a kubernetes.deployment already knows its facts and how to probe them.

components:
  api:
    type: kubernetes.deployment
    depends:
      - on: rds
  rds:
    type: aws.rds_instance

Simulate in CI¶

A scenario injects facts and states the expected conclusion: "rds is down and api is crash-looping. Blame rds, not api." mgtt simulate checks the engine agrees. It needs no cluster and no credentials. If someone drops a dependency or a health rule, the PR fails before the model can mislead anyone in an incident.

$ mgtt simulate --all
  api crash-loop, rds healthy              ✗ FAILED
    expected: root_cause=api path=[nginx, api] eliminated=[rds, frontend]
    actual:   root_cause=api path=[nginx, api] eliminated=[frontend]

Diagnose at 3am¶

mgtt diagnose runs read-only probes against the live system and keeps narrowing until one failure chain is left:

Root cause: rds
Scenario:   rds.stopped → api.crashed → nginx.degraded
Probes run: 7/20   Time: 3.2s/5m0s

It doesn't need the person who built the system, because the model already holds that knowledge. When a probe is refused (RBAC, IAM) or times out, the report says Cannot rule out: rds. It never counts what it couldn't see as healthy.

AI agents get the same engine over MCP. They can run the diagnosis within your limits (read-only providers only, a cap on probes per incident, whether write probes pause or are refused). They can also draft the model and its scenarios from your providers' real vocabulary, checking both before you review. See AI agents.

Retrospective¶

$ mgtt incident end --emit-scenario
wrote scenarios/inc-20261004-0814-001.yaml — passes; commit it to keep this diagnosis under test

The facts mgtt saw during the incident become a scenario. After you commit it, every future PR must still produce that diagnosis.

Start: Quick start · How it works