Grafana MCP

Official Grafana Labs MCP server — dashboards, Prometheus, Loki, alerts and incidents in your LLM client

Free· Free and open source (Apache 2.0). You still need a Grafana instance — OSS Grafana is free; Grafana Cloud has a free tier plus paid Pro/Advanced/Enterprise plans.MCP Servers

Visit website →

Best for

SRE, platform and observability teams who already run Grafana (OSS or Cloud) and want Claude, Cursor or an internal agent to investigate incidents, query Prometheus/Loki, and read dashboards conversationally.

Skip if

Teams without a Grafana deployment, users who want a hosted no-config chatbot, or anyone looking for a general-purpose analytics tool unconnected to Grafana datasources.

Grafana MCP (mcp-grafana) is the official Model Context Protocol server maintained by Grafana Labs that lets LLM clients like Claude Desktop, Cursor, VS Code, Windsurf and other MCP-compatible assistants query and operate a Grafana instance in natural language. Written in Go and distributed as an Apache-2.0 binary, uvx package, or Docker image, it exposes Grafana's REST API surface as a curated set of MCP tools — searching and reading dashboards, pulling panel queries and JSONPath fragments, rendering panels as PNG for the model to see, listing and inspecting datasources, running PromQL against Prometheus, LogQL against Loki, and SQL against InfluxDB, ClickHouse, Elasticsearch, Snowflake, CloudWatch, Athena, Graphite and Quickwit with Grafana's macro substitution applied. On the operations side it wraps alert rules and notification policies, Grafana Incident (create incidents, add activity, update severity), Grafana OnCall (list schedules, current shift, users), Sift investigations for automated error-pattern detection, annotations, snapshots, and provisioning-repository validation. Recent releases add Agent Observability tooling for LLM conversation tracking and Grafana Assistant integration. The server speaks stdio for a single desktop client or SSE / Streamable HTTP for multi-client deployments; write-capable tools are gated behind an --enabled-tools allowlist and a --disable-write read-only mode, so an SRE team can safely hand a scoped service-account token to an agent without granting a blank cheque on production dashboards.

Editor's take

The reference implementation for wiring an LLM into an observability stack. The scope is huge — dashboards, alerts, incidents, on-call, Sift, plus first-class Prometheus and Loki — but the read-only flag and per-tool allowlist mean you can actually let an agent loose on prod without losing sleep. If you run Grafana, install this before you build anything custom.

— The AI Tool Bible editorial team

Pros

✅ Official Grafana Labs project, actively developed with 3k+ GitHub stars and Apache-2.0 licensing
✅ Broad coverage: dashboards, alerts, incidents, on-call, Sift, annotations, snapshots, plus native query support for Prometheus, Loki, and eight SQL/timeseries backends
✅ Panel PNG rendering lets vision-capable LLMs actually see charts, not just JSON
✅ Three transports (stdio, SSE, Streamable HTTP) cover single-user desktop and multi-client server deployments
✅ Per-tool enable list and --disable-write flag make it safe to hand to autonomous agents against production Grafana
✅ Works out of the box with Claude Desktop, Cursor, VS Code and any MCP-spec client via uvx or Docker

Cons

⚠️ Requires Grafana 9.0+ and a service account token — no value without an existing Grafana deployment
⚠️ Many powerful tools are disabled by default and must be explicitly enabled, which is safer but adds config friction
⚠️ Broad tool surface can flood a model's context window; you often need to curate which tools are exposed per assistant
⚠️ PromQL/LogQL/SQL responses are raw datasource output — the LLM still has to reason about large result sets, which burns tokens fast
⚠️ Not a hosted service: you run and secure the process yourself, and network reachability to Grafana is your problem

Use cases

Incident investigation from Claude DesktopConversational PromQL and LogQL queryingDashboard search and summarisationPanel screenshot analysis by vision LLMsAlert rule and notification policy reviewOn-call schedule lookupsSift automated error-pattern triageCursor / VS Code observability copilotMulti-tenant SSE server for internal agentsGrafana Incident timeline updates from an agent