Grafana’s API, restructured for LLM attention spans
It turns your Grafana instance into a toolkit for MCP clients, letting an LLM query metrics, patch dashboards, and check who’s on call without writing your own glue code.

What it does
This is an official Grafana adapter that exposes the platform’s sprawling API as Model Context Protocol tools. An LLM client like Claude or Cursor can search dashboards, run PromQL or LogQL queries, manage alert rules, browse on-call schedules, and even create incidents. It speaks to your Grafana instance and whatever datasources you have wired up—Prometheus, Loki, ClickHouse, Snowflake, Athena, and others.
The interesting bit
The maintainers clearly understand that LLMs have short attention spans: they provide JSONPath extraction and compact dashboard summaries so you don’t have to dump a megabyte of JSON into the context window just to find a panel title. Most sharp tools—like direct panel execution, admin operations, and many datasource queries—ship disabled by default behind feature flags.
Key highlights
- Covers the full Grafana surface area: dashboards, alerting, OnCall schedules, incidents, Sift investigations, and RBAC admin.
- Supports querying a wide range of datasources through Grafana: Prometheus, Loki, CloudWatch, ClickHouse, Snowflake, Athena, Elasticsearch/OpenSearch, InfluxDB, and Graphite.
- Context-window-aware dashboard tools: fetch summaries, extract specific properties with JSONPath, or patch targeted changes instead of passing full JSON.
- Many tool groups (InfluxDB, ClickHouse, CloudWatch, Graphite, Athena, Snowflake, Elasticsearch, admin, panel execution) are disabled by default and must be explicitly enabled.
Caveats
- Requires Grafana 9.0 or later; some datasource operations will misbehave on older versions because of missing API endpoints.
- Large dashboards can still consume significant context window space if you fetch them whole.
- The README notes that the feature list is informational and “does not represent a roadmap or commitment to future features.”
Verdict
Worth a look if you want your editor’s AI assistant to actually interrogate your observability stack rather than hallucinate about it. Skip it if you don’t run Grafana or you’re uncomfortable letting an LLM near production alert rules.
Frequently asked
- What is grafana/mcp-grafana?
- It turns your Grafana instance into a toolkit for MCP clients, letting an LLM query metrics, patch dashboards, and check who’s on call without writing your own glue code.
- Is mcp-grafana open source?
- Yes — grafana/mcp-grafana is open source, released under the Apache-2.0 license.
- What language is mcp-grafana written in?
- grafana/mcp-grafana is primarily written in Go.
- How popular is mcp-grafana?
- grafana/mcp-grafana has 3.4k stars on GitHub and is currently accelerating.
- Where can I find mcp-grafana?
- grafana/mcp-grafana is on GitHub at https://github.com/grafana/mcp-grafana.