Grafana MCP — Install & Live Demo

Name: Grafana MCP MCP Server
Author: grafana

Why use it

Key features

Prometheus query / range query / labels / metadata
Loki LogQL query + label discovery
Tempo trace search
Dashboard list/get; panel-level data
Alerting rules + alert instances + silences
OnCall schedules and shifts
Sift investigations (Grafana Cloud)
Read-only mode supported

Live Demo

What it looks like in practice

mcp-grafana.replay ▶ ready

0/0

Install

Pick your client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "mcp-grafana": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GRAFANA_URL",
        "-e",
        "GRAFANA_API_KEY",
        "mcp/grafana",
        "-t",
        "stdio"
      ],
      "env": {
        "GRAFANA_URL": "https://your-grafana.example.com",
        "GRAFANA_API_KEY": "${GRAFANA_API_KEY}"
      }
    }
  }
}

Open Claude Desktop → Settings → Developer → Edit Config. Restart after saving.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "mcp-grafana": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GRAFANA_URL",
        "-e",
        "GRAFANA_API_KEY",
        "mcp/grafana",
        "-t",
        "stdio"
      ],
      "env": {
        "GRAFANA_URL": "https://your-grafana.example.com",
        "GRAFANA_API_KEY": "${GRAFANA_API_KEY}"
      }
    }
  }
}

Cursor uses the same mcpServers schema as Claude Desktop. Project config wins over global.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "mcp-grafana": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GRAFANA_URL",
        "-e",
        "GRAFANA_API_KEY",
        "mcp/grafana",
        "-t",
        "stdio"
      ],
      "env": {
        "GRAFANA_URL": "https://your-grafana.example.com",
        "GRAFANA_API_KEY": "${GRAFANA_API_KEY}"
      }
    }
  }
}

Click the MCP Servers icon in the Cline sidebar, then "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "mcp-grafana": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GRAFANA_URL",
        "-e",
        "GRAFANA_API_KEY",
        "mcp/grafana",
        "-t",
        "stdio"
      ],
      "env": {
        "GRAFANA_URL": "https://your-grafana.example.com",
        "GRAFANA_API_KEY": "${GRAFANA_API_KEY}"
      }
    }
  }
}

Same shape as Claude Desktop. Restart Windsurf to pick up changes.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "mcp-grafana",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "GRAFANA_URL",
        "-e",
        "GRAFANA_API_KEY",
        "mcp/grafana",
        "-t",
        "stdio"
      ]
    }
  ]
}

Continue uses an array of server objects rather than a map.

~/.config/zed/settings.json

{
  "context_servers": {
    "mcp-grafana": {
      "command": {
        "path": "docker",
        "args": [
          "run",
          "-i",
          "--rm",
          "-e",
          "GRAFANA_URL",
          "-e",
          "GRAFANA_API_KEY",
          "mcp/grafana",
          "-t",
          "stdio"
        ]
      }
    }
  }
}

Add to context_servers. Zed hot-reloads on save.

claude mcp add mcp-grafana -- docker run -i --rm -e GRAFANA_URL -e GRAFANA_API_KEY mcp/grafana -t stdio

One-liner. Verify with claude mcp list. Remove with claude mcp remove.

Use Cases

Real-world ways to use Grafana MCP

Build an incident timeline from Grafana data

👤 SREs / on-call ⏱ ~10 min intermediate

When to use: Pager just fired; you want a coherent timeline before joining the call.

Prerequisites

Grafana service-account token — Grafana → Administration → Service accounts; Editor role is enough for queries

Flow

What's firing

Grafana: list firing alerts in the last 30 minutes for service=checkout.✓ Copied

→ list_alerts returns 1+ alerts with timestamps
Pull the metric

For each alert, run a Prometheus range query for the underlying metric over the last 1h. Note the breach time.✓ Copied

→ query_prometheus_range returns time series
Pull logs at breach

Loki: logs for service=checkout, level=error, [breach_time-2m, breach_time+2m]. Top patterns.✓ Copied

→ Log lines clustered by signature
Compose timeline

Build a concise timeline: alert fired → metric breach → top 3 error log patterns. Markdown for Slack.✓ Copied

→ Timeline ready to paste

Outcome: Coherent incident timeline assembled before you join the call.

Pitfalls

Loki query too broad → blows token budget — Always include service= label; limit time window aggressively
Prometheus query has step too fine — Use step=15s or 30s for 1h windows

Combine with: sentry

Explain what a Grafana dashboard actually shows

👤 New team member onboarding ⏱ ~15 min beginner

When to use: You inherited a dashboard and want to understand it without clicking each panel.

Flow

Find it

Grafana: search dashboards titled 'Checkout SLO'.✓ Copied

→ Dashboard UID returned
Inspect panels

Get the dashboard. For each panel, summarize: title, datasource, what query it runs, what threshold (if any).✓ Copied

→ Per-panel explanation; query strings included

Outcome: Documented understanding of the dashboard ready for onboarding wiki.

Pitfalls

Variables substituted weirdly — Mention $service/$env interpolation is dashboard-time, not MCP-time

Audit alert rules for noise

👤 SRE leads ⏱ ~30 min intermediate

When to use: Quarterly hygiene — find rules that fire constantly or never.

Flow

Pull rules

Grafana: list all alert rules in folder Production.✓ Copied

→ Rule list with UID + expression
Get fire history

For each, count alert instances in last 30 days. Flag rules with >100 (noisy) or 0 (dormant).✓ Copied

→ Per-rule counts + flagged set
Recommend

For each flagged rule, propose a fix: tighten threshold, add for: 5m, or delete dormant.✓ Copied

→ Concrete YAML changes per rule

Outcome: Concrete cleanup plan with PR-ready YAML.

Pitfalls

Counting instances ≠ counting unique alerts — Group by labels; count distinct firing periods

Combine with: github

Combinations

Pair with other MCPs for X10 leverage

mcp-grafana + sentry

Cross-correlate Sentry errors with metric/log breaches

Sentry: get the spike in errors for OrderService. Grafana: pull Prometheus + Loki around the same window.✓ Copied

mcp-grafana + github

Open PR with alert-rule cleanups

Grafana: audit alert rules. GitHub: open a PR replacing the noisy rules with tightened thresholds.✓ Copied

mcp-grafana + filesystem

Save incident timeline to a Markdown post-mortem

Build a timeline from Grafana, save to /pm/2026-05-incident.md.✓ Copied

Tools

What this MCP exposes

Tool	Inputs	When to call	Cost
list_datasources	(none)	First call, learn what's available	1 API call
query_prometheus	datasource_uid, query, time?	Instant metric value	1 call
query_prometheus_range	datasource_uid, query, start, end, step	Time-series investigation	1 call
query_loki_logs	datasource_uid, expr, start, end, limit?	Pull logs	1 call
search_dashboards	query: str	Find a dashboard	1 call
get_dashboard	uid: str	Inspect panels	1 call
list_alert_rules	folder?	Audit / discover	1 call
list_alert_instances	state?	What's firing now	1 call
list_oncall_schedules	(none)	Who's on call	1 call

Cost & Limits

What this costs to run

API quota: Bound by your Grafana instance limits; Cloud has org-level QPS caps
Tokens per call: 300–8000 (range queries can be large)
Monetary: Free OSS; Grafana Cloud bills normally
Tip: Always pass step= on range queries; use Loki limit=100 unless you know the volume

Security

Permissions, secrets, blast radius

Minimum scopes: service-account with Viewer or Editor

Credential storage: Service account token in env var

Data egress: Only your Grafana URL

Never grant: Admin role unless you need rule mutation

Use --read-only flag if Claude shouldn't be able to mutate alerts/silences
Service account tokens don't expire by default — rotate quarterly

Troubleshooting

Common errors and fixes

401 Unauthorized

Token wrong or service account disabled

Verify: curl -H 'Authorization: Bearer $TOKEN' $URL/api/health

Prometheus query timeout

Reduce time range or step

Loki 'parse error'

LogQL needs label filter on left of pipeline; e.g. {service="x"} |= "err"

Dashboard not found

UID changed after import; search by title first

Alternatives

Grafana MCP vs others

Alternative	When to use it instead	Tradeoff
Direct Prometheus MCP	Prometheus only, no Grafana	No dashboards, no alerts UI integration
Datadog MCP	You're on Datadog	Different observability stack
Loki MCP standalone	Logs only	Less surface

More

Resources

📖 Read the official README on GitHub

🐙 Browse open issues

🔍 Browse all 400+ MCP servers and Skills