Turn Claude Code telemetry into org-wide adoption dashboards. Track cost per user, slice metrics by department and manager, and correlate AI usage with Jira/Tempo productivity data.
Built on top of Claudefana (the core OTEL stack for Claude Code observability).
- Adoption tracking — active users, adoption rate, new vs returning users over time
- Cost attribution — total spend, cost per user/session/commit, broken down by model
- Org-level slicing — filter everything by department, manager, or rollup manager (via Microsoft Graph)
- Jira/Tempo correlation — resolved issues, story points, and worklog hours alongside AI metrics
- Developer scorecard — full-width table ranking users by cost, output, and productivity
- Org hierarchy tree — interactive rollup → manager → user drill-down
- Per-user deep dive — drill into any user's sessions, costs, tool usage, and output over time
- Prometheus alerts — 10 built-in alerts for usage limits, burn rate, cache efficiency, and exporter health
Two Grafana dashboards with 60+ panels total, 3 cascading filters (rollup_name → manager_name → user_email), and automatic user discovery from Claude Code telemetry.
Claude Code ──OTLP──▶ OTEL Collector ──▶ Prometheus ──▶ Grafana
│ ▲ ▲
▼ │ │
Loki │ │
│ │
graph-enrichment ──┘ │
(Microsoft Graph) │
│
jira-tempo-exporter ─────┘
(Jira REST + Tempo API)
6 services total. The core 4 come from claudefana; this repo adds the 2 enterprise exporters and replaces the dashboards.
- Clone both repos as siblings:
git clone https://github.com/JuanjoFuchs/claudefana.git
git clone https://github.com/JuanjoFuchs/claudefana-enterprise.git- Configure credentials:
cd claudefana-enterprise
cp .env.example .env
# Edit .env with your Jira and Azure AD credentials- Start everything:
docker compose -f docker-compose.enterprise.yaml up -d- Open Grafana at http://localhost:3000
Dashboards auto-populate as Claude Code users generate telemetry. The exporters discover users automatically from Prometheus — no user list to maintain.
| Variable | Description |
|---|---|
JIRA_URL |
Your Atlassian instance URL (e.g. https://your-org.atlassian.net) |
JIRA_API_TOKEN |
API token from id.atlassian.com |
STORY_POINTS_FIELD |
Your Jira custom field ID for story points (check your Jira instance) |
PROJECT_FILTER |
Optional comma-separated project keys to scope collection |
COLLECT_TEAMS |
Set to true to collect Tempo team memberships (expensive: 1+N API calls) |
| Variable | Description |
|---|---|
GRAPH_TENANT_ID |
Azure AD tenant ID |
GRAPH_CLIENT_ID |
App registration client ID |
GRAPH_CLIENT_SECRET |
App registration client secret |
Requires an Azure AD App Registration with User.Read.All (Application) permission and admin consent.
Kustomize manifests in k8s/base/ deploy all 6 services to any Kubernetes cluster.
docker build -t your-registry/claudefana/jira-tempo-exporter:latest ./jira-tempo-exporter
docker build -t your-registry/claudefana/graph-enrichment-exporter:latest ./graph-enrichment-exporter
docker push your-registry/claudefana/jira-tempo-exporter:latest
docker push your-registry/claudefana/graph-enrichment-exporter:latestkubectl create namespace claudefana
kubectl -n claudefana create secret generic claudefana-jira \
--from-literal=JIRA_URL=https://your-org.atlassian.net \
--from-literal=JIRA_API_TOKEN=your-token \
--from-literal=PROJECT_FILTER=
kubectl -n claudefana create secret generic claudefana-graph \
--from-literal=AZURE_TENANT_ID=your-tenant-id \
--from-literal=AZURE_CLIENT_ID=your-client-id \
--from-literal=AZURE_CLIENT_SECRET=your-secretEdit k8s/base/deployments.yaml:
- Replace
your-registry/with your container registry URL - Set
STORY_POINTS_FIELDto match your Jira instance
Optionally uncomment storageClassName in k8s/base/pvcs.yaml to match your cluster.
kubectl apply -k k8s/base/An example ingress (AWS ALB) is provided at k8s/base/ingresses.yaml.example. Copy it to a Kustomize overlay and customize for your ingress controller and hostnames.
For org-specific configuration (SSO, real hostnames, registry URLs), create an overlay in a separate repo or directory:
your-deploy-repo/
└── k8s/
├── kustomization.yaml # References base resources + dashboard configMapGenerator
├── ingresses.yaml # Real hostnames
└── patches/
├── grafana-sso.yaml # OAuth env vars
├── images.yaml # Registry URL overrides
└── pvcs.yaml # Storage class override
Note: The base's
configMapGeneratorreferences../../dashboards/*.json, which works when runningkubectl apply -k k8s/base/directly but triggers a Kustomize security error when consumed as a sub-resource from another directory. To work around this, overlays should reference the base's individual resource files and provide their ownconfigMapGeneratorfor the dashboard JSON:# overlay kustomization.yaml resources: - path/to/claudefana-enterprise/k8s/base/namespace.yaml - path/to/claudefana-enterprise/k8s/base/configmaps.yaml - path/to/claudefana-enterprise/k8s/base/deployments.yaml - path/to/claudefana-enterprise/k8s/base/services.yaml - path/to/claudefana-enterprise/k8s/base/pvcs.yaml - ingresses.yaml configMapGenerator: - name: grafana-dashboard-json files: - path/to/claudefana-enterprise/dashboards/claude-code-dashboard.json - path/to/claudefana-enterprise/dashboards/claude-code-user-explorer.json patches: - path: patches/grafana-sso.yamlCross-directory references also require
--load-restrictor=LoadRestrictionsNone:kubectl kustomize k8s/ --load-restrictor=LoadRestrictionsNone | kubectl apply -f -
Both exporters query Prometheus to discover which users have Claude Code telemetry, then fetch external data only for those users. No user lists to configure — adoption grows automatically.
- Graph Enrichment — resolves each user's department, job title, manager, and rollup manager (CxO direct report) from Microsoft Graph. Exports a
user_infogauge that acts as a label bridge for PromQL joins. - Jira/Tempo — pulls resolved/created issues, story points, and Tempo worklogs scoped to Claude Code users. Enables cost-per-ticket and AI-time-vs-billable-time analysis.
All org-level dashboard queries use the same PromQL pattern:
sum by (department)(
claude_code_cost_usage_USD_total
* on(user_email) group_left(department)
user_info{rollup_name=~"$rollup_name"}
)
- Claudefana core stack (cloned as sibling directory)
- Docker Compose or Kubernetes cluster
- Jira Cloud/Server + Tempo (for Jira/Tempo exporter)
- Azure AD with App Registration (for Graph enrichment)
- Grafana plugin:
equansdatahub-tree-panel(auto-installed in K8s; manual for Docker Compose)
MIT