MCP Grafana

What grafana/mcp-grafana does

grafana/mcp-grafana gives AI agents access to Grafana dashboards, alerting, incidents, annotations, datasources, and observability data through MCP. It can inspect dashboards at different levels of detail, from compact metadata to complete JSON, and can extract panel queries or selected properties with JSONPath. Dashboard creation, updates, patching, folders, snapshots, annotations, and deeplink generation are also supported.

The server covers operational investigation as well as configuration work. Agents can inspect or modify alert rules, routing policies, contact points, time intervals, and silences. Incident tools support creation, retrieval, and timeline notes, while OnCall tools expose alert groups, current on-call users, and shift details.

How it works

The server connects to a Grafana URL using a service account token and exposes Grafana operations as MCP tools. Datasource authentication remains configured in Grafana; SQL queries, for example, are sent through Grafana datasource plugins rather than handling database credentials in the MCP server.

Dashboard retrieval includes context-management options. Use summaries for planning, JSONPath properties for targeted inspection, and full dashboard retrieval only when the complete model is needed. Panel queries can include template-variable substitutions, and panel query execution supports custom time ranges and variable overrides when enabled.

Several tool families are disabled by default and must be enabled with the corresponding --enabled-tools value: runpanelquery, examples, influxdb, sql, cloudwatch, cloudlogging, and graphite.

Setup and configuration

A standard setup uses uvx to launch mcp-grafana from an MCP client's server configuration:

{
  "mcpServers": {
    "grafana": {
      "command": "uvx",
      "args": ["mcp-grafana"],
      "env": {
        "GRAFANA_URL": "https://your-instance.grafana.net",
        "GRAFANA_SERVICE_ACCOUNT_TOKEN": "your-token"
      }
    }
  }
}

The same configuration pattern can point to a local Grafana installation. The project also supports Docker, a binary, and Helm deployment. Full functionality requires Grafana 9.0 or newer; some datasource operations depend on API endpoints unavailable in earlier versions.

Tools and capabilities

• Search, summarize, inspect, create, update, patch, and version dashboards.
• Retrieve panel queries, datasource details, annotations, snapshots, folders, and documentation pages.
• Query Prometheus, Loki, InfluxDB, SQL datasources, CloudWatch, Google Cloud Logging, and Graphite when their tool groups are enabled.
• Analyze Loki label strategies, error patterns, slow Tempo requests, and differences between traces.
• Manage alert rules, routing, silences, incidents, annotations, and OnCall information.
• Generate dashboard, panel, and Explore deeplinks.

Supported datasource examples include Prometheus, Loki, ClickHouse, CloudWatch, Elasticsearch, OpenSearch, Snowflake, Athena, MySQL, PostgreSQL, and MSSQL. The Google Cloud Logging tools require the Google Cloud Logging datasource plugin version 1.8.0 or later and Grafana 11.2 or newer.

Limitations and notes

Large dashboards can use substantial context-window space, so targeted summary and property tools are preferable when full JSON is unnecessary. Running panel queries and several datasource query families are opt-in rather than enabled by default. Creating incidents can notify many people and should be confirmed before execution. Deleting annotations is permanent, and updating a silence requires care with its original start time to preserve its identifier.