# Grafana MCP — Dashboards, Alerts & OnCall Tools

> Grafana MCP server connects your MCP client to Grafana so agents can search dashboards, query panels, and automate ops tasks with a service account token.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Install `uv` and add to your MCP config:
   ```json
   {
     "mcpServers": {
       "grafana": {
         "command": "uvx",
         "args": ["mcp-grafana"],
         "env": {
           "GRAFANA_URL": "http://localhost:3000",
           "GRAFANA_SERVICE_ACCOUNT_TOKEN": "<your token>"
         }
       }
     }
   }
   ```
2. Restart your MCP client.
3. Ask the agent to “search dashboards” and fetch a dashboard summary.

## Intro

Grafana MCP server connects your MCP client to Grafana so agents can search dashboards, query panels, and automate ops tasks with a service account token.

- **Best for:** SRE/ops teams who want agents to navigate Grafana dashboards and automate common read/write tasks safely
- **Works with:** Grafana 9.0+, service account token, MCP clients via uvx or Docker/Helm installs
- **Setup time:** 12 minutes

## Practical Notes

- Requires Grafana 9.0+ for full functionality (README requirement)
- Some tool categories are disabled/limited by flags; plan least-privilege RBAC scopes (README RBAC section)

## A Least-Privilege MCP Pattern for Observability

Treat your Grafana token like production credentials.

Suggested steps:

1. Create a dedicated Grafana service account for MCP.
2. Start with **read-only** operations (dashboard search, summaries).
3. Only then enable write tools (dashboard patch/update) when you’re comfortable with the audit trail.

To keep agent context small:

- Prefer summary tools over full dashboard JSON.
- Use JSONPath-style property extraction to pull only what you need.

This makes Grafana automation scalable: the agent can answer “what changed?” and “what’s broken?” without flooding the context window.

### FAQ

**Q: Do I need a Grafana token?**
A: Yes. The README uses `GRAFANA_SERVICE_ACCOUNT_TOKEN` for authentication.

**Q: Can I run it without Docker?**
A: Yes. The README shows a uvx-based quick start and also Docker/Helm options.

**Q: Why does context size matter?**
A: Dashboard JSON can be huge; use summaries and property extraction to avoid token blowups.

## Source & Thanks

> Source: https://github.com/grafana/mcp-grafana
> License: Apache-2.0
> GitHub stars: 2,989 · forks: 353

---

<!-- ZH -->

## 快速使用

1. 安装 `uv` 并加入 MCP 配置：
   ```json
   {
     "mcpServers": {
       "grafana": {
         "command": "uvx",
         "args": ["mcp-grafana"],
         "env": {
           "GRAFANA_URL": "http://localhost:3000",
           "GRAFANA_SERVICE_ACCOUNT_TOKEN": "<your token>"
         }
       }
     }
   }
   ```
2. 重启 MCP 客户端。
3. 让 Agent 先“搜索仪表盘”，再拉一个 dashboard summary 验证可用。

## 简介

Grafana MCP server 让 MCP 客户端直连你的 Grafana：用 service account token 搜索仪表盘、读取/更新内容、运行查询并辅助运维自动化，适合把排障步骤变成可重复的“工具链”。

- **适合谁：** 希望让 Agent 更高效地导航 Grafana 仪表盘，并安全自动化常见运维读写任务的团队
- **可搭配：** Grafana 9.0+、service account token、通过 uvx 或 Docker/Helm 接入 MCP 客户端
- **准备时间：** 12 分钟

## 实战建议

- README 要求 Grafana 9.0+ 才能完整工作
- 工具类别可用 flags 开关；按 README 的 RBAC 建议规划最小权限 scopes

## 面向可观测性的最小权限模式

把 Grafana token 当成生产凭据对待。

建议步骤：

1. 为 MCP 单独创建 service account。
2. 先只开只读能力（搜索仪表盘、获取摘要）。
3. 等审计链路与误操作风险可控后，再逐步开放写入类工具（patch/update）。

为了控制上下文体积：

- 优先用 summary，而不是拉全量 dashboard JSON。
- 用 property/JSONPath 式的提取只取必要字段。

这样 Agent 才能在不爆上下文的前提下回答“发生了什么 / 哪里坏了”。

### FAQ

**必须有 Grafana token 吗？**
答：需要，README 使用 `GRAFANA_SERVICE_ACCOUNT_TOKEN` 作为认证方式。

**不装 Docker 也能跑吗？**
答：可以，README 提供 uvx 快速启动，也有 Docker/Helm 选项。

**为什么要管上下文大小？**
答：dashboard JSON 很大；用 summary 与字段提取避免 token 爆炸。

## 来源与感谢

> Source: https://github.com/grafana/mcp-grafana
> License: Apache-2.0
> GitHub stars: 2,989 · forks: 353


---
Source: https://tokrepo.com/en/workflows/grafana-mcp-dashboards-alerts-oncall-tools
Author: MCP Hub