MCP Victoriametrics

What VictoriaMetrics-Community/mcp-victoriametrics MCP server does

VictoriaMetrics-Community/mcp-victoriametrics MCP server connects an MCP client to a VictoriaMetrics deployment and exposes most of the read-only capabilities available in VMUI. It is intended for monitoring, observability, and troubleshooting workflows where an agent needs to inspect metrics or understand VictoriaMetrics configuration.

The server can query time-series data, enumerate metrics and labels, retrieve label values, and export series. It also supports investigation of alerting and recording rules, active alerts, instance parameters, metric cardinality, and usage statistics. Query-focused functions include analysis, tracing, formatting, and explanation. Configuration-oriented functions cover relabeling rules, downsampling, and retention policies.

An embedded copy of VictoriaMetrics documentation can be searched without network access. VictoriaMetrics Cloud is supported in addition to single-node and clustered VictoriaMetrics installations.

How it works

The process translates MCP requests from a compatible client into requests against the configured VictoriaMetrics instance and its documented APIs. Its tools are primarily read-only, so the material describes inspection and analysis rather than data mutation. Results depend on the permissions and behavior of the connected VictoriaMetrics deployment, as well as the MCP client and model handling the responses.

The server can run in different transport modes, including stdio and Streamable HTTP. In Streamable HTTP mode, the root endpoint provides setup instructions and a tool-inspection UI. A hosted test endpoint at https://play-mcp.victoriametrics.com/mcp connects to the public VictoriaMetrics Playground and is intended for MCP clients rather than ordinary browser visits.

Setup and configuration

VictoriaMetrics-Community/mcp-victoriametrics MCP server requires access to a VictoriaMetrics or VictoriaMetrics Cloud instance. Installation options include downloading a release binary, running the published container image, building from source with Go, or using the VictoriaMetrics MCP Helm chart.

Configuration is supplied through environment variables. VM_INSTANCE_ENTRYPOINT identifies the root URL of a vmsingle or vmselect endpoint, while VM_INSTANCE_TYPE specifies the deployment type. The configuration also documents VMC_API_KEY for VictoriaMetrics Cloud access. Transport and listener settings include MCP_SERVER_MODE and MCP_LISTEN_ADDR; the Docker example uses the sse mode and port 8080. The README states that Go 1.26 or newer is needed when building from source.

A Docker-based HTTP setup can use the published image ghcr.io/victoriametrics/mcp-victoriametrics. Replace the example instance settings with the endpoint and type for the target deployment. For a local client using stdio, use the downloaded binary or a locally built binary according to that client’s MCP configuration format.

Tools and capabilities

Supported workflows include:

• Run metric queries and inspect returned time-series data.
• List metrics, labels, label values, and complete series.
• Examine alerting rules, recording rules, and alerts.
• Review instance parameters and metric cardinality.
• Inspect metric usage statistics.
• Trace, explain, analyze, and prettify queries.
• Test relabeling rules and investigate downsampling or retention settings.
• Search the bundled VictoriaMetrics documentation offline.

The README notes that clients with graph-rendering support may be able to draw query results.

Limitations and notes

The documented API surface is described as almost entirely read-only; the supplied material does not establish write or mutation operations. Access to useful data requires a reachable VictoriaMetrics or VictoriaMetrics Cloud instance, except when using the public test deployment. The public endpoint is tied to the VictoriaMetrics Playground and is intended for evaluation rather than ordinary browser access.

The quality of responses depends on the MCP client and the model in use. The README also notes that the server can be combined with other observability or documentation-search MCP servers, but those are separate components and are not included here.