Grafana MCP Server vs Datadog MCP Server

When to pick Grafana MCP Server

• Your investigation starts with a Grafana dashboard. Grafana MCP Server can extract panel queries and datasource details, fetch saved dashboard versions, and return a compact overview through get_dashboard_summary before you request the full JSON.
• You need to query Prometheus or Loki through existing Grafana datasources. The server runs PromQL and LogQL, discovers metric names and labels, and retrieves Loki log patterns. You can inspect the query behind a panel and then investigate its underlying data.
• You maintain dashboards as part of development work. Dashboard patches support targeted changes without submitting the complete dashboard JSON. Provisioning tools can also validate a repository file at a branch or commit before applying it.
• Your incident workflow uses Grafana Incident and OnCall. Tools cover incident searches and updates, schedule details, alert groups, and current on-call users. Access depends on the service account's permissions and the relevant plugin actions.

When to pick Datadog MCP Server

• Your logs, metrics, traces, and incidents are already in Datadog. Datadog MCP Server exposes those observability resources to agents, so an investigation can query the data your team already collects there.
• You want a managed connection without installing a local server. The client connects to Datadog's regional HTTPS endpoint and authenticates with OAuth or an access token. Choose the endpoint that matches your Datadog site.
• Your work includes Datadog error tracking, security signals, or RUM. These capabilities are available as toolsets alongside dashboards and alerting; the connection URL determines which toolsets load.
• You want to limit the tools presented to an agent for a particular task. Datadog's URL-based toolset selection lets you select capabilities for the connection instead of exposing every available group.

Main differences

• Deployment choices differ. Grafana's quick start launches a local process, with Docker, binary, and Helm options also documented; Grafana Cloud offers a hosted endpoint. Datadog leads with a managed server and also publishes a local binary and a Claude Code plugin.
• Connection configuration differs. Grafana supports stdio, SSE, and streamable HTTP. Datadog's documented managed setup uses an HTTP connection to a regional HTTPS URL, so the site hostname is part of the client configuration.
• Authentication follows each platform. Grafana's quick start uses a service account token, with RBAC permissions and scopes controlling accessible operations and resources. Datadog's managed connection uses OAuth or an access token.
• Tool selection happens in different places. Grafana uses startup flags such as --enabled-tools and category exclusions such as --disable-oncall. SQL, InfluxDB, CloudWatch, and panel-query execution are among its opt-in groups. Datadog selects toolsets through the connection URL.
• Query scope follows where your data is configured. Grafana exposes connected datasources, including SQL queries routed through Grafana plugins; SQL datasource credentials remain in that configuration and are never seen by the MCP server. Datadog's managed endpoint exposes Datadog observability data and product toolsets, including traces, RUM, and security signals.

Setup

For Grafana, install uv, then configure your MCP client to run uvx mcp-grafana. Set GRAFANA_URL to your Grafana instance and GRAFANA_SERVICE_ACCOUNT_TOKEN to a service account token. Assign permissions and resource scopes for the tools you intend to use; use your instance URL for Grafana Cloud.

For Datadog on US1, run claude mcp add --transport http datadog https://mcp.datadoghq.com/v1/mcp in Claude Code, then authorize with OAuth or an access token. Replace the hostname for another Datadog site; EU uses mcp.datadoghq.eu. Clients supporting .mcp.json can configure the same endpoint with type http.