@neuralbridge/observability-mcp
v1.0.0
Published
MCP server for read-only observability: query Prometheus metrics, Loki logs, Grafana dashboards and Alertmanager alerts over their HTTP APIs.
Maintainers
Readme
@neuralbridge/observability-mcp
Read-only MCP server for observability stacks. It queries tools the user already runs, over their HTTP APIs, and never installs or changes anything:
- Prometheus — PromQL instant and range queries, firing/pending alerts.
- Loki — LogQL log queries.
- Grafana — search dashboards, fetch a dashboard by uid.
- Alertmanager — list alerts.
Configuration
Each service is enabled by its URL in the environment. Authentication is optional and provided per service as a bearer token or a username/password pair. Secrets stay in the environment and are never printed in tool output.
| Service | URL | Auth (optional) |
|---|---|---|
| Prometheus | PROMETHEUS_URL | PROMETHEUS_TOKEN or PROMETHEUS_USERNAME + PROMETHEUS_PASSWORD |
| Loki | LOKI_URL | LOKI_TOKEN or LOKI_USERNAME + LOKI_PASSWORD |
| Grafana | GRAFANA_URL | GRAFANA_TOKEN or GRAFANA_USERNAME + GRAFANA_PASSWORD |
| Alertmanager | ALERTMANAGER_URL | ALERTMANAGER_TOKEN or ALERTMANAGER_USERNAME + ALERTMANAGER_PASSWORD |
NB_OBS_TIMEOUT_MS sets the per-request timeout in milliseconds (default 30000,
maximum 120000).
Use obs_detect to see which services are configured.
Tools
All tools are read-only.
obs_detectobs_prometheus_query,obs_prometheus_range,obs_prometheus_alertsobs_loki_queryobs_grafana_search,obs_grafana_dashboardobs_alertmanager_alerts
Build and test
npm install
npm run build
npm test