Prometheus
Let the agent run PromQL against your Prometheus, list metrics and check scrape targets.
The Prometheus MCP server runs instant and range PromQL queries, lists and describes metrics, and shows scrape targets. Prometheus' query API can't write, so the server is read-only by design.
Already running Grafana?
The Grafana server queries your Prometheus datasources too, with dashboards and Loki on top. Use this server when you want Prometheus directly, without Grafana.
| Setting | Value |
|---|---|
| Runs | In your network, next to Prometheus |
| Endpoint URL | https://<your hostname>/mcp |
| Transport | Streamable HTTP |
| Authentication | Bearer token, through Supergateway (the server has no key of its own) |
Run it in HTTP mode
docker run -d --name mcp-prometheus -p 8080:8080 \
-e PROMETHEUS_URL=http://prometheus.monitoring.svc:9090 \
-e PROMETHEUS_MCP_SERVER_TRANSPORT=http \
-e PROMETHEUS_MCP_BIND_HOST=0.0.0.0 \
-e PROMETHEUS_MCP_BIND_PORT=8080 \
ghcr.io/pab1it0/prometheus-mcp-server:latestIf Prometheus itself needs credentials, add PROMETHEUS_TOKEN, or PROMETHEUS_USERNAME and PROMETHEUS_PASSWORD.
The server prints its MCP endpoint (usually /mcp) when it starts. A Helm chart
is also available.
Put a key and https in front
supergateway --streamableHttp "http://mcp-prometheus:8080/mcp" \
--outputTransport streamableHttp --stateful --port 8000 --apiKey "$MCP_API_KEY"Expose Supergateway over https. See Expose a self-hosted server. A self-hosted sremcp in the same network can use the server's address directly.
Onboard in sremcp
| Field | Value |
|---|---|
| Display name / Slug | Prometheus (prod) / prometheus-prod |
| Endpoint URL | https://mcp-prometheus.example.com/mcp |
| Transport | Streamable HTTP |
| Authentication | Bearer token, Secret = MCP_API_KEY |
Approve tools
All of them are reads: execute_query, execute_range_query, list_metrics, get_metric_metadata,
get_targets and health_check.
Try it
- What's the p99 latency of http_server_requests for checkout over the last 6 hours, and when did it change?
- Which scrape targets are down?
- Compare CPU usage of the payments pods with this time yesterday.