Quick answer: Grafana MCP through MCP Express lets Claude check firing alerts, review alert rules, and create or delete silences directly — so you can triage and quiet noisy alerts in plain English instead of logging into Grafana's UI. Today, MCP Express's Grafana integration covers alerting and silence management specifically, not dashboard building or metric querying. Setup takes about 5 minutes.
It's 11 PM. Your phone lights up — a Grafana alert. You already know what it is before you open it: the same disk-usage warning that's fired four times this week on a client's staging server that nobody's touched since Tuesday.
You could silence it. That means opening a laptop, logging into Grafana, finding the right alert rule, and clicking through the silence dialog — for a warning you've already triaged three times.
Multiply that across three clients' Grafana instances, and alert fatigue stops being an annoyance. It becomes a real risk: the night you're too tired to check "just one more notification" is the night the alert that actually matters gets lost in the noise.
The real cost of alert fatigue when you're managing someone else's infrastructure
Alert fatigue isn't a Grafana problem — it's a triage problem. Every alerting system will eventually fire on something that isn't urgent: a flaky check, a known maintenance window, a threshold that's slightly too sensitive. The fix isn't fewer alerts. It's faster, lower-friction triage.
For a freelancer or small ops team managing Grafana across multiple clients, that friction compounds. Each client's instance means its own login, its own dashboard layout, its own muscle memory for where the silence button lives. Ten minutes of "let me just check what's firing" turns into logging into three separate Grafana instances before you've even had coffee.
What Grafana MCP actually does today (and what it doesn't)
Worth being upfront about this, since a lot of what you'll find written about "Grafana MCP" covers dashboard generation and PromQL querying — the open-source Grafana MCP server does support that. MCP Express's Grafana integration is scoped specifically to alerting and silence management:
- List Alert Rules — see all configured alert rules
- Get Firing Alerts — pull currently firing alerts, with optional label filters
- List Silences — see what's currently silenced
- Create Silence — silence a specific alert rule by UID
- Delete Silence — remove an existing silence
If you need Claude to build dashboards or run raw metric queries, that's outside this integration's current scope. What it's genuinely good at is the exact loop that eats your evenings: check what's firing, decide if it matters, silence what doesn't.
Prerequisites
- Claude (Claude web or Claude Desktop)
- MCP Express account (free tier, no credit card)
- A Grafana instance with alert rules configured, and API access credentials
Estimated time: 5 minutes from signup to first query.
Setting Up Grafana MCP in 4 Steps
Step 1: Add the Grafana Integration
In MCP Express, go to MCP Servers → New MCP Server.
Name it per client or environment — "Client-Prod-Grafana" keeps things clear once you're managing more than one instance.
Select Grafana from the integrations list and enter your instance URL and token key.
Turn on List Alert Rules, Get Firing Alerts, List Silences, Create Silence, and Delete Silence — or just the subset relevant to how you work.
Add a tool description so Claude knows when to reach for it: "Check and manage Grafana alerts and silences for [client]'s production environment."
Describing Grafana MCP for AIStep 3: Connect Claude to your Grafana MCP Server
You have two options, depending on how your team uses Claude:
The easy way — OAuth (recommended)
In Claude, go to Customize → Connectors and click Add new Connector. Give it a name and set the URL to:
https://api.mcp-express.com/gateway/mcp
Claude will redirect you to MCP Express to complete the OAuth flow. Log in, grant access, and select the MCP server you just configured. No config files, no CLI, no Node.js required — anyone on your team can connect in the same way.
The manual way — MCP Express CLI
If you prefer the manual config route, we've covered it step by step in our connection guide. It takes under 5 minutes.
What This Actually Looks Like Day to Day
- "What's currently firing on the client-prod server?" — Claude pulls active alerts with their labels, so you see what's urgent without opening a browser tab.
- "Silence the disk-usage alert on staging until Monday." — Claude creates the silence directly, no dialog boxes.
- "What alert rules do we have configured for the payments service?" — a quick audit without hunting through Grafana's rule list manually.
- "Are there any silences about to expire?" — catch one before it lapses and starts paging you again for something you already knew about.
Manual Grafana Triage vs. Grafana MCP
| | |
|---|
Checking firing alerts | Log in, navigate to Alerting, filter | Ask Claude directly |
Silencing a known non-issue | Find the rule, open silence dialog, set duration | One sentence, done |
Managing multiple clients' instances | Separate login and layout per instance | Same conversation, any configured instance |
Available from | Wherever you can open Grafana's UI | Any device signed into Claude |
None of this replaces actually investigating an alert that matters — that judgment call is still yours. What it removes is the friction between "I know this is noise" and "this is actually silenced."
Frequently Asked Questions
- What is Grafana MCP?
It's a way for an AI model like Claude to interact with your Grafana instance's alerting system through the Model Context Protocol — checking firing alerts, listing rules, and managing silences via plain-English requests instead of the Grafana UI.
- Does MCP Express's Grafana integration support dashboards or metric queries?
Not currently — it's scoped to alerting and silence management: listing alert rules, checking firing alerts, and creating/deleting silences.
- Can Claude silence an alert permanently by mistake?
Silences require a specific alert rule UID and duration you set — Claude isn't guessing, it's executing exactly what you ask it to create.
- Can I use this across multiple clients' Grafana instances?
Yes — configure a separate MCP server per instance, and manage all of them from the same Claude conversation.
- Do I need Grafana admin access to set this up?
You need API credentials with permission to read alert rules and manage silences — check with whoever administers the Grafana instance if you're not the admin.
Quiet the Noise, Not the Signal
You've got Claude checking firing alerts and managing silences across your Grafana instances — without the nightly login-and-click routine.
Connect Grafana free → No credit card required, takes about 5 minutes.
