Grafana MCP Server

mcp-grafana is the official Model Context Protocol server from Grafana Labs that exposes a Grafana instance (self-hosted or Grafana Cloud) to MCP-compatible clients like Claude Desktop, Cursor, and VS Code Copilot. It gives AI assistants programmatic access to query metrics and logs, search and manage dashboards, manage alert rules, work with Incident and Sift, and generate deeplinks back into the Grafana UI.

The server ships 40+ tools organized into roughly 16 categories: Dashboards, Datasources, Prometheus, Loki, Alerting, Incidents, OnCall, Sift, Navigation, Annotations, Rendering, Admin/RBAC, plus disabled-by-default integrations for InfluxDB, ClickHouse, CloudWatch, Elasticsearch, Snowflake, Athena, and Graphite. It supports stdio, SSE, and streamable-http transports, runs via uvx, Docker, a Go binary, or a Helm chart, and authenticates with a Grafana service account token.

Because it is maintained by Grafana Labs directly, it tracks the Grafana HTTP API closely, supports both Grafana OSS (9.0+) and Grafana Cloud with the same configuration, and includes operational niceties like --disable-write for read-only mode, OTLP tracing, Prometheus metrics, and granular per-category tool toggles.

Tools

Prerequisites

• Grafana 9.0+ (self-hosted or Grafana Cloud)
• A Grafana service account token with appropriate RBAC permissions. The built-in Editor role works for most read/write operations.
• One of: uvx, Docker, the Go binary, or the Helm chart.

Install

The recommended path is uvx:

uvx mcp-grafana

Or pull the Docker image:

docker run --rm -i \
  -e GRAFANA_URL=http://localhost:3000 \
  -e GRAFANA_SERVICE_ACCOUNT_TOKEN=<token> \
  grafana/mcp-grafana -t stdio

Claude Desktop / Cursor config

Using uvx:

{
  "mcpServers": {
    "grafana": {
      "command": "uvx",
      "args": ["mcp-grafana"],
      "env": {
        "GRAFANA_URL": "http://localhost:3000",
        "GRAFANA_SERVICE_ACCOUNT_TOKEN": "<token>"
      }
    }
  }
}

Using Docker:

{
  "mcpServers": {
    "grafana": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-e", "GRAFANA_URL",
        "-e", "GRAFANA_SERVICE_ACCOUNT_TOKEN",
        "grafana/mcp-grafana", "-t", "stdio"
      ],
      "env": {
        "GRAFANA_URL": "http://localhost:3000",
        "GRAFANA_SERVICE_ACCOUNT_TOKEN": "<token>"
      }
    }
  }
}

Environment variables

• GRAFANA_URL: Grafana instance URL (e.g. http://localhost:3000 or https://myinstance.grafana.net)
• GRAFANA_SERVICE_ACCOUNT_TOKEN: Service account token (recommended)
• GRAFANA_USERNAME / GRAFANA_PASSWORD: Basic auth alternative
• GRAFANA_ORG_ID: Optional org id for multi-org setups
• GRAFANA_EXTRA_HEADERS: Additional headers as JSON

Useful flags

• -t stdio|sse|streamable-http: Transport mode (default stdio)
• --enabled-tools datasource,prometheus,loki,...: Enable extra categories like influxdb, clickhouse, cloudwatch, elasticsearch, snowflake, athena, graphite, admin, runpanelquery
• --disable-write: Run in read-only mode (disables update_dashboard, create_incident, create_annotation, alert rule mutations, Sift investigations)
• --metrics and --metrics-address :9090: Expose Prometheus metrics
• --debug / --log-level debug: Verbose logging

• Ask an agent to query PromQL or LogQL and summarize anomalies during an incident, then create a Grafana Incident with an activity log of what it found.
• Triage on-call alerts by listing current OnCall users, fetching the active alert group, and running Sift find_error_pattern_logs against the relevant Loki datasource.
• Have an LLM search dashboards, fetch panel queries, and propose an updated dashboard JSON, then apply it via update_dashboard.
• Generate deeplinks to specific dashboards or Explore views and drop them into Slack or a postmortem doc with the right time range pre-filled.
• Annotate dashboards programmatically from a deploy pipeline or chat workflow using create_annotation.

• "Query Prometheus for the p99 latency of the checkout service over the last 1h and chart the top 5 routes."
• "Search Loki for error patterns in the payments namespace in the last 30 minutes and summarize the top three."
• "Who is currently on call for the SRE team, and is there an open alert group right now?"
• "Find the Production Overview dashboard, list its panel queries, and generate a deeplink scoped to the last 6 hours."
• "Create a Grafana Incident titled 'Elevated 5xx on api-gateway' with a summary of the recent Sift analysis."

• Maintained by Grafana Labs as the official MCP server, with regular releases and tight coverage of the Grafana HTTP API.
• Broad surface area: 40+ tools across dashboards, Prometheus, Loki, alerting, incidents, OnCall, Sift, annotations, RBAC, and rendering.
• Flexible deployment: uvx, Docker, Go binary, or Helm chart, with stdio, SSE, and streamable-http transports.
• Operational controls like --disable-write, per-category tool toggles, OTLP tracing, and a Prometheus /metrics endpoint.

• Several integrations (InfluxDB, ClickHouse, CloudWatch, Elasticsearch, Snowflake, Athena, Graphite, admin, run_panel_query) are disabled by default and must be enabled with --enabled-tools.
• Requires a Grafana service account token with carefully scoped RBAC; misconfigured permissions silently break tools.
• Some advanced features (Incidents, OnCall, Sift) depend on Grafana Cloud or specific plugins and will not work on a bare OSS Grafana install.

• grafana/loki-mcp: Focused MCP server for Grafana Loki only.
• grafana/tempo-mcp-server: MCP server for Grafana Tempo traces.
• prometheus-community Prometheus MCP servers: Community MCP servers that talk directly to Prometheus without going through Grafana.