Grafana Agent

What you can do

A few things this template does well out of the box.

• Walk a dashboard panel-by-panel and summarise what each one is showing right now
• Inspect alert rules: what conditions trigger them, current state, recent history
• Find dashboards that reference a specific data source or metric — useful before deprecating either
• Pull a Prometheus / Loki / Tempo query result and explain the trend
• Manage on-call routing through Grafana IRM (when the IRM toolset is enabled)

How it works

Three steps to go from template to a live chat.

MCP servers used

The endpoints this template connects to by default. You can swap any of them in Agent Studio.

Getting your access token

A quick walkthrough for the credential this template needs.

1. 1**Grafana Cloud users**: the default URL `https://mcp.grafana.com/mcp` is correct. OAuth 2.1 is the auth model — you'll be prompted to authorise the connection on first use.
2. 2**Service-account fallback**: if OAuth isn't set up, create a Grafana service account at **Administration → Service accounts** with the roles you need (Viewer for read-only, Editor for managing dashboards). Generate a token and paste it into **Access token**.
3. 3**OSS Grafana users**: the hosted MCP is Cloud-only. Run the OSS server locally: `docker run -p 8000:8000 grafana/mcp-grafana --grafana-url https://your-grafana --grafana-api-key YOUR_KEY` and replace the URL with `http://localhost:8000`.
4. 4**Toolset scoping**: append `?toolsets=dashboards,alerting` (etc.) to limit which tools the agent sees. Reduces context, speeds up responses.
5. 5Send *"List the dashboards in this org"* to confirm.

Try these prompts

Copy one into the studio to see the agent in action.

• Walk me through the "Production API" dashboard panel by panel. What is each panel showing right now?

• Which alert rules are firing right now in this org? Who owns them?

• Find every dashboard that queries the `prometheus-prod` data source. Are any of them broken?

• Pull p95 latency for the `checkout` service from Prometheus for the last 6 hours and explain the trend.

• Compare error rate on `api-gateway` between us-east-1 and eu-west-1 over the last week.

System prompt

The default instructions the model starts with. Edit it any time inside Agent Studio.

You are a Grafana observability assistant connected via the official Grafana Cloud MCP server.

Use the available tools to:
- List, inspect and walk through dashboards panel-by-panel
- Read alert rule definitions and current state; surface what's firing, what's pending, and recent state changes
- Query Prometheus, Loki and Tempo data sources directly through Grafana
- Inspect data source health and find dashboards that depend on a specific data source
- Manage on-call schedules and incidents (when Grafana IRM toolset is enabled)

Operating principles:
- When summarising a dashboard, group panels by what they tell the on-call: golden signals first (latency, errors, saturation), business metrics second
- For alert rule investigations, lead with "is this firing right now and why" before going into rule definitions
- For PromQL/LogQL queries, explain the query intent in one sentence before running it
- Never modify a dashboard or alert rule without explicit user confirmation
- If a panel uses a data source you can't reach, say so — don't guess what the panel would show