ArgoCD MCP — Anwendungsfälle, Installieren & Live-Demo

Warum nutzen

Hauptfunktionen

List/inspect ArgoCD applications across projects
Show sync status, health, and last-sync info
Diff live cluster state vs desired git state
Trigger sync (with prune/force options) when authorized
Read app manifests without cluster kubectl access

Live-Demo

In der Praxis

argocd.replay ▶ bereit

0/0

Installieren

Wählen Sie Ihren Client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "argocd": {
      "command": "uvx",
      "args": [
        "mcp-for-argocd"
      ]
    }
  }
}

Öffne Claude Desktop → Settings → Developer → Edit Config. Nach dem Speichern neu starten.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "argocd": {
      "command": "uvx",
      "args": [
        "mcp-for-argocd"
      ]
    }
  }
}

Cursor nutzt das gleiche mcpServers-Schema wie Claude Desktop. Projektkonfiguration schlägt die globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "argocd": {
      "command": "uvx",
      "args": [
        "mcp-for-argocd"
      ]
    }
  }
}

Klicken Sie auf das MCP-Servers-Symbol in der Cline-Seitenleiste, dann "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "argocd": {
      "command": "uvx",
      "args": [
        "mcp-for-argocd"
      ]
    }
  }
}

Gleiche Struktur wie Claude Desktop. Windsurf neu starten zum Übernehmen.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "argocd",
      "command": "uvx",
      "args": [
        "mcp-for-argocd"
      ]
    }
  ]
}

Continue nutzt ein Array von Serverobjekten statt einer Map.

~/.config/zed/settings.json

{
  "context_servers": {
    "argocd": {
      "command": {
        "path": "uvx",
        "args": [
          "mcp-for-argocd"
        ]
      }
    }
  }
}

In context_servers hinzufügen. Zed lädt beim Speichern neu.

claude mcp add argocd -- uvx mcp-for-argocd

Einzeiler. Prüfen mit claude mcp list. Entfernen mit claude mcp remove.

Anwendungsfälle

Praxisnahe Nutzung: ArgoCD

Find ArgoCD apps that have drifted out of sync

👤 Platform / SRE teams running GitOps ⏱ ~20 min intermediate

Wann einsetzen: Weekly: which apps in which clusters are OutOfSync or Degraded, and why?

Voraussetzungen

ArgoCD API token with read scope — argocd account generate-token --account <read-only-user>
ArgoCD server URL — ARGOCD_SERVER=argocd.my.company.com

Ablauf

List apps with status

List all ArgoCD apps. For each: name, project, sync status, health status, last sync time.✓ Kopiert

→ Full inventory
Focus on drift

Filter to apps with syncStatus != 'Synced' or health != 'Healthy'. Rank by time since last sync.✓ Kopiert

→ Problem apps list
Diff a specific app

For app <name>, show the diff between desired (git) and live. What resources are out of sync?✓ Kopiert

→ Resource-level diff

Ergebnis: A weekly drift report identifying which apps need attention and why.

Fallstricke

Drift caused by a legitimate runtime-only resource (e.g. HPA scaled replicas) — Configure ignoreDifferences in the Application spec to exclude fields that change at runtime

Kombinieren mit: notion

Review and apply a sync for a specific application

👤 DevOps engineers promoting changes ⏱ ~15 min intermediate

Wann einsetzen: A PR merged to main; ArgoCD shows the app as OutOfSync pending your approval.

Voraussetzungen

Token with sync permission on the target project — Role with applications, sync, <project>/* allowed

Ablauf

Inspect the pending change

For app <name>, show the diff. What resources change, what's the blast radius?✓ Kopiert

→ Concrete diff
Check the source commit

What commit SHA does the desired state point at? Show the GitHub commit message and PR.✓ Kopiert

→ Commit + PR context
Sync with prune=false first

Trigger sync for <name> with prune=false, dryRun=false. Wait for completion, show final status.✓ Kopiert

→ Sync Succeeded; health Healthy

Ergebnis: A deploy with prior-review, not a surprise; deletion-by-prune only happens explicitly.

Fallstricke

Sync with prune=true can delete resources you still need — Always start with prune=false; only enable prune when you've confirmed no manually-created resources exist in the namespace

Kombinieren mit: github

Emergency rollback to a previous git revision

👤 SREs during a bad-deploy incident ⏱ ~15 min advanced

Wann einsetzen: A deploy went out, errors spiked, and you need to roll back to last known good.

Ablauf

Get current + history

For app <name>, show current revision and the last 5 deployed revisions with timestamps.✓ Kopiert

→ History table
Pick the rollback target

What was the revision 2 syncs ago? Confirm its SHA matches the git commit before the incident started.✓ Kopiert

→ Target SHA identified
Rollback

Sync <name> to revision <SHA> with prune=false. Watch until Healthy. Post in Teams when done.✓ Kopiert

→ App reverted + notification sent

Ergebnis: A clean rollback with an audit trail, typically under 5 minutes.

Fallstricke

Rolling back only the app but DB migration already ran forward — DB migrations need their own rollback plan; sometimes 'roll forward' with a fix is safer than rolling back code against mutated schema

Kombinieren mit: sentry · ms-teams

Build a cross-cluster app inventory

👤 Platform teams managing many clusters ⏱ ~20 min intermediate

Wann einsetzen: You run ArgoCD in a hub-spoke model and want a flat list: 'which apps run where'.

Ablauf

List clusters ArgoCD manages

List all clusters registered in ArgoCD with their name and server URL.✓ Kopiert

→ Cluster inventory
Map apps to clusters

For each app, show its destination cluster and namespace. Group by cluster.✓ Kopiert

→ Per-cluster app list
Flag unusual placements

Any app deployed in an unexpected cluster (e.g. 'prod-only' app in a dev cluster)? Flag with reason.✓ Kopiert

→ Anomaly list

Ergebnis: An up-to-date map of what runs where, useful for audits and capacity planning.

Fallstricke

Stale cluster entries — secrets removed but cluster still listed — Periodically verify each cluster is reachable; remove stale ones

Kombinationen

Mit anderen MCPs für 10-fache Wirkung

argocd + github

Cross-reference Argo sync with the PR that caused it

For app <name>, show the current revision SHA; fetch the GitHub PR that introduced it and summarize the change.✓ Kopiert

argocd + sentry

Correlate an Argo sync with a post-deploy error spike

App <name> synced at 14:02; did Sentry errors spike after that? If yes, show the top issue introduced.✓ Kopiert

argocd + ms-teams

Post sync/rollback events to a Teams channel

After any manual sync of app <name>, post a message to Teams #deploys with revision SHA and who triggered it.✓ Kopiert

Werkzeuge

Was dieses MCP bereitstellt

Werkzeug	Eingaben	Wann aufrufen	Kosten
list_applications	project?, selector?	Inventory apps	free
get_application	name	Detailed state of a specific app	free
get_application_diff	name, revision?	Preview what a sync would change	free
sync_application	name, revision?, prune?, dryRun?, resources?	Apply pending changes	free (triggers cluster work)
get_application_resource_tree	name	Inspect the live resources an app owns	free
list_clusters		Cross-cluster inventory	free

Kosten & Limits

Was der Betrieb kostet

API-Kontingent: Bounded by your ArgoCD server capacity; practical limit ~10 req/s
Tokens pro Aufruf: App list: 500–3000 tokens. Diff: up to 5000.
Kosten in €: Free — ArgoCD is open source, you only pay for the infra hosting it.
Tipp: Cache app list; use selectors (project, label) to narrow queries rather than listing everything.

Sicherheit

Rechte, Secrets, Reichweite

Minimale Scopes: read on applications, clusters for read-only sessions

Credential-Speicherung: ARGOCD_AUTH_TOKEN and ARGOCD_SERVER in env

Datenabfluss: Calls to your ArgoCD server only

Niemals gewähren: admin sync on all projects without reason account:update

Sync permission can deploy arbitrary manifests — treat it like production write access.
Use project-scoped RBAC policies instead of cluster-admin for automation accounts.

Fehlerbehebung

Häufige Fehler und Lösungen

401 Unauthenticated

ARGOCD_AUTH_TOKEN expired or revoked. Regenerate via argocd account generate-token.

Prüfen: argocd account get-user-info --auth-token $ARGOCD_AUTH_TOKEN

403 Permission denied on sync

Role lacks applications, sync, <project>/* permission. Update AppProject or account role.

Sync stuck in Progressing for >10 min

Usually a stuck resource — check resource-tree for a failing hook or a stuck controller; may need manual intervention in-cluster.

Diff shows unexpected drift on every sync

Controllers mutating fields at runtime. Add ignoreDifferences to the Application spec for those fields.

Alternativen

ArgoCD vs. andere

Alternative	Wann stattdessen	Kompromiss
Flux MCP	You use Flux instead of ArgoCD for GitOps	Different GitOps engine; CLI-first model
Kubernetes MCP (raw kubectl)	You want direct cluster access without GitOps abstraction	Imperative — defeats the purpose of GitOps for managed apps

Mehr

Ressourcen

📖 Offizielle README auf GitHub lesen

🐙 Offene Issues ansehen

🔍 Alle 400+ MCP-Server und Skills durchsuchen