Grafana MCP Server for Claude

Content

The Grafana MCP server connects Claude and other MCP-capable clients to a Grafana instance. It gives an assistant a structured way to query observability data, inspect dashboards, work with alerting and incident workflows, and generate accurate deeplinks back into Grafana.

This is a strong fit for teams that already use Grafana as their operations hub. Instead of asking an assistant to guess URLs or invent PromQL, users can give Claude access to Grafana-backed tools with explicit authentication, transport, and RBAC boundaries.

The official quick-start path uses uvx mcp-grafana, but Grafana also documents Docker, downloaded binary, source build, and Helm deployment options. Grafana 9.0 or later is required for full functionality.

Features

• Query Prometheus metrics through Grafana datasources.
• Query Loki logs and LogQL-backed metrics.
• Inspect datasource metadata, labels, metric names, and query examples.
• Search dashboards and fetch dashboard summaries, properties, panel queries, and datasource details.
• Generate Grafana deeplinks for dashboards, panels, Explore, time ranges, and datasource-specific views.
• List and manage Grafana alert rules, notification policies, contact points, and related alerting configuration when permissions allow.
• Work with Grafana Incident, Sift investigations, and Grafana OnCall resources when those products and permissions are available.
• Enable or disable tool groups so the MCP surface matches the team's risk tolerance and Grafana setup.
• Run locally over stdio with uvx, Docker, or a binary, or run as an HTTP server with SSE or streamable HTTP transports.

Use Cases

• Ask Claude to investigate a recent latency spike using Grafana metrics and narrowed time windows.
• Search dashboards and retrieve only the panels or JSON paths needed for a specific incident review.
• Generate Grafana Explore links for a Prometheus or Loki query without relying on hand-built URLs.
• Review alert rule state, routing, and notification policy context during an incident.
• Use Grafana Incident or OnCall context alongside logs and metrics when triaging production issues.
• Give an assistant read-only observability context while keeping dashboard and alert writes disabled or out of scope.

Installation

Claude Code

1. Install uv and confirm uvx is available.
2. Create a Grafana service account token with only the permissions needed for the tool groups you plan to use.
3. Add the MCP server with your Grafana URL and token:

claude mcp add grafana --env GRAFANA_URL=https://myinstance.grafana.net --env GRAFANA_SERVICE_ACCOUNT_TOKEN=YOUR_SERVICE_ACCOUNT_TOKEN -- uvx mcp-grafana

1. Restart or refresh the MCP client session.
2. Ask Claude to list available Grafana tools or run a narrow dashboard or datasource lookup.

Claude Desktop

1. Open the Claude Desktop MCP configuration file.
2. Add the grafana server configuration shown below.
3. Replace the placeholder URL and token with your Grafana instance URL and service account token.
4. Restart Claude Desktop and test with a narrow read-only query first.

Configuration

{
  "grafana": {
    "command": "uvx",
    "args": ["mcp-grafana"],
    "env": {
      "GRAFANA_URL": "https://myinstance.grafana.net",
      "GRAFANA_SERVICE_ACCOUNT_TOKEN": "<your service account token>"
    }
  }
}

Examples

Inspect dashboard context

Ask Claude to find the production API dashboard, summarize its panels, and show which Prometheus and Loki datasources it uses.

Find dashboards related to the production API and summarize the panels and datasources.

Investigate a metric spike

Use the Prometheus tools with a specific datasource, time window, and metric name instead of broad exploratory queries.

For the last 30 minutes, query the production Prometheus datasource for API p95 latency and error rate.

Review log patterns

Use Loki against a bounded time range and service label when looking for common errors.

Check Loki logs for the checkout service over the last 15 minutes and group recurring error messages.

Generate a Grafana deeplink

Ask Claude to create a Grafana Explore link for a specific datasource query and time range.

Create a Grafana Explore link for this Loki query over the incident window.

Security

• Prefer service account token authentication over username and password for automation.
• Grant only the RBAC actions and scopes required for your enabled tools. Use datasource, dashboard, folder, and team scopes where possible instead of broad organization-wide access.
• Start with read-only investigation workflows. Add write-capable dashboard, alerting, incident, or annotation tools only when the operator has a clear approval process.
• Keep tokens in MCP environment configuration, not in prompts, chat history, or checked-in files.
• Review generated queries before running them against large production datasources, especially when they include wide time ranges or unbounded label selectors.

Troubleshooting

Claude cannot start the server

Confirm uvx is installed and available in the same environment that launches your MCP client. If you use Docker, a downloaded binary, or Helm instead, update the client configuration to match that runtime.

Authentication fails

Verify GRAFANA_URL points at the correct Grafana instance and that GRAFANA_SERVICE_ACCOUNT_TOKEN belongs to a service account with the required RBAC permissions.

Queries return permission errors

Check the service account role and scopes for the datasource, dashboard, folder, alerting, or incident resource being accessed. A token can authenticate successfully and still lack permission for a specific tool.

Responses are too large

Use dashboard summaries, specific dashboard properties, narrower time windows, and more precise label filters before fetching full dashboard JSON or broad log results.