Netdata#Netdata

What netdata/netdata#Netdata does

netdata/netdata#Netdata connects MCP-compatible AI assistants to a Netdata agent. Through that connection, an assistant can inspect observability data such as time-series metrics, logs, alerts, and live system details. Live information can include processes, services, containers, virtual machines, and network connections.

The integration is intended for infrastructure discovery, monitoring investigation, reporting, and root-cause analysis. It can expose system, application, and hardware metrics collected by Netdata, including data from CPUs, memory, disks, network interfaces, databases, web servers, containers, IPMI sensors, GPUs, and temperature sensors.

How it works

The MCP server is built into Netdata, so installing a separate server package is not required. An assistant connects to the Netdata agent selected by its configuration. The visibility available to the assistant depends on that connection point:

• A standalone or child node provides visibility into that node.
• A parent node can provide visibility across its connected child nodes.
• Netdata Cloud support is planned but is not available in this integration preview.

Netdata exposes MCP over the dashboard's HTTP port. The /mcp endpoint supports streamable HTTP, and the same endpoint can use Server-Sent Events with the SSE transport option. Clients that only support stdio can use nd-mcp, which forwards stdio communication to the Netdata MCP endpoint.

Setup and configuration

Install Netdata on the system that will provide the MCP connection, then configure the AI assistant with the path to nd-mcp, an API key, and the Netdata MCP WebSocket URL. The bridge is included with Netdata installations; common locations include /usr/bin/nd-mcp, /usr/sbin/nd-mcp, and platform-specific installation directories.

The agent generates an API key for the integration at startup. It is stored in Netdata's data directory, commonly under /var/lib/netdata/mcp_dev_preview_api_key or the corresponding static-installation path. The key authorizes access to sensitive functions such as logs and live system information.

Clients supporting HTTP or SSE can connect directly to the agent's /mcp endpoint with a Bearer authorization header and do not need the bridge. The official @modelcontextprotocol/remote-mcp CLI can be used to test either transport. For stdio-only clients, configure the bridge as the MCP command. Claude Desktop uses its developer configuration, Claude Code can use a project .mcp.json or the claude mcp add command, and Cursor uses its MCP settings.

Tools and capabilities

The integration provides access to:

• Infrastructure and node discovery, including hardware, operating system, virtualization, streaming, and connection details.
• Discovery of metrics collected by the connected Netdata installation.
• System, application, hardware, container, and network observability data.
• Alerts and logs for investigation.
• Live information about processes, services, containers, virtual machines, and network connections.

Go, Node.js, and Python stdio-to-WebSocket bridges are also included for environments that need an alternative to nd-mcp. Their intended functionality is equivalent; each bridge contains a build script, and the Go bridge also includes a Windows build script.

Limitations and notes

The assistant only sees nodes reachable through the Netdata instance to which it connects. Parent-node connections can cover multiple child nodes, while child and standalone connections are limited to one node. Clients without HTTP or SSE support need a stdio-to-network bridge. The integration is available through Netdata agents; the planned Netdata Cloud connection is not yet available.