← All docs

Connecting Grafana

Metrics, logs, traces, and alert rules, over Grafana's MCP server.

Why MCP rather than the Grafana API

Your metrics might be in Prometheus, your logs in Loki, your traces in Tempo — three query languages and three APIs. Grafana's MCP server fronts all of them with one tool surface, so Incident Copilot speaks one protocol instead of several, and picking up a datasource you add later needs no change here.

What you need

  1. The MCP endpoint URL of your Grafana MCP server (commonly https://<your-grafana>/api/plugins/grafana-mcp/mcp, or the /mcp path of a standalone mcp-grafana deployment)
  2. A Grafana service account token with Viewer access to the datasources you want consulted

Set it up

  1. Connectors → Add connector → Grafana
  2. Paste the MCP endpoint URL and the service account token
  3. Click Test connection

Read the test result

A successful test shows the server name, the protocol version, and — importantly — the list of tools that server actually advertises. Incident Copilot uses those tools by name, and different Grafana MCP builds expose different sets. If the list looks unexpectedly short, incidents from this connector will arrive with thinner context, and that is the moment to notice it rather than during an outage.

What is read, and what is not

Read-only, always. The Grafana connector queries datasources, alert rules, and incidents. It has no method that writes anything back to Grafana — silencing an alert, editing a dashboard, or acknowledging an incident are all things Incident Copilot deliberately cannot do.