Mcp on Nicole van der Hoeven

When to use Grafana Assistant vs MCP vs GCX: Grafana MCP

Wed, 19 Aug 2026 09:00:00 +0100

Three AI tools people confuse: Assistant, MCP, and GCX. Last time I talked about the brain — Assistant. Now hand number one: the MCP server, the easy hand.

Grafana MCP — Model Context Protocol — lets any AI client reach into Grafana. The magic part: you bring your own brain.

Using MCP requires two pieces:

The client is your AI app — the brain you bring: claude.ai, ChatGPT, Cursor.
The server is the door into Grafana that exposes the tools.

There are two ways to get that server. The easiest is the Grafana Cloud hosted one — already running, auth built in, just connect and go. But you can also self-host the open-source one on your own infra.

So when do you reach for Grafana MCP?

When you’re using hosted chat tools with no terminal — claude.ai, ChatGPT, your phone. No shell means no CLI, so MCP is the only way in.
When you want to bolt Grafana onto a tool you already use — already in Cursor or Claude? Just add the server. No new workflow.
When you want the best discovery out of the box — the tools describe themselves up front, so even smaller or cheaper models just know what’s available. No trial-and-error.

One honest catch: those tool definitions can eat some context if you load everything — but modern hosts fix that with lazy tool search.

MCP is deliberately curated — it doesn’t expose every corner of Grafana. So what do you do when you need the esoteric stuff? That’s exactly what the power hand, GCX, is for — next up.

Transcript

Three AI tools people confuse: Assistant, MCP, GCX. Last time, I talked about the brain — Assistant. Now hand number one: the MCP server, the easy hand.

Grafana MCP — Model Context Protocol — lets any AI client reach into Grafana. The magic part: you bring that AI client - you bring your own brain.

Using MCP requires two pieces. The client is your AI app — the brain you bring: claude.ai, ChatGPT, Cursor. The server is the door into Grafana that exposes the tools.

Two ways to get that server: the easiest way is to use the Grafana Cloud hosted one — already running, auth built in, just connect and go. But you can also self-host the open-source one on your own infra.

So when do you reach for Grafana MCP?

When you’re using hosted chat tools with no terminal — claude.ai, ChatGPT, your phone. No shell means no CLI, so MCP is the only way in.
When you want to bolt Grafana onto a tool you already use — already in Cursor or Claude? Just add the server. No new workflow.
When you want the best discovery out of the box — the tools describe themselves up front, so even smaller or cheaper models just know what’s available. No trial-and-error.

One honest catch: those tool definitions can eat some context if you load everything — but modern hosts fix that with lazy tool search.

But MCP is deliberately curated — it doesn’t expose every corner of Grafana. So what do you do when you need the esoteric stuff? That’s exactly what the power hand is for. Next up is GCX.

Resources

AI Week Day 2: Building (Grafana AI Week — Tuesday)

Tue, 28 Jul 2026 09:00:00 +0100

Day two of AI Week is all about building — the AI tools that help you write software with confidence. Two big announcements went generally available: gcx (the Grafana CLI) and the Grafana MCP server. Both connect your coding agents — Claude Code, Cursor, and friends — to your Grafana telemetry. The MCP server gives agents opinionated, ready-made tools for the common cases (querying metrics and logs, reading and updating dashboards); gcx is the more flexible, less opinionated option, built for agents from the ground up with read-only guardrails and its own bundle of agent skills. Both work for OSS and Grafana Cloud. The point: when an agent writes your code, it shouldn’t be guessing — it should be looking at how your system actually behaves in production. That’s telemetry-driven development.

Transcript

It’s day two of Grafana Labs’ AI Week, and Tuesday is all about building — the AI tools that help you build software with confidence.

Today we have two big announcements: gcx and the Grafana MCP server are both now generally available.

Both of them connect your coding agents — think Claude Code or Cursor — to your Grafana telemetry. So what’s the difference?

The Grafana MCP server gives your agent opinionated, ready-made tools for the most common use cases: querying your metrics and logs, reading and updating dashboards, that kind of thing. You can run it yourself, or use the hosted MCP server with your Grafana Cloud stack — so yes, it works for both OSS and Grafana Cloud.

gcx is the Grafana CLI, and it’s the more flexible, less opinionated option. Agents can explore it and build their own workflows, and it was designed for agents from the ground up, with read-only guardrails so your agent can’t touch anything you didn’t approve. It works with Cloud stacks and self-managed instances, and it even ships with its own bundle of agent skills.

Why does this matter? Because when an agent writes your code, it shouldn’t be guessing — it should be looking at how your system actually behaves in production. That’s telemetry-driven development.

Head to grafana.ai for everything we’re shipping this week, and I’ll see you tomorrow for day three.

CLI and MCP Basics for AI Agents (The Context Window #04)

Wed, 03 Jun 2026 18:00:00 +0100

MCP vs CLI: which one should your AI agent actually use? In this episode of The Context Window — Grafana’s livestream series on AI in observability — I’m joined by Tiffany Jernigan and three of the engineers and PMs behind Grafana’s agent tooling: Ben Sully (who wrote the Grafana MCP server), and Ward Bekker and Dafydd Thomas from the team building gcx. We get into the two main ways an AI agent can reach the outside world — the Model Context Protocol (MCP) and good old command-line interfaces — and look at when each one makes sense, the tradeoffs around context windows, tool sprawl, token cost, and reliability. We walk through Grafana’s cloud and open-source MCP servers along with gcx, the Grafana Cloud CLI built for agents, do a live side-by-side demo of both creating an SLO, and talk about how we benchmark them against each other with o11y-bench.

Timestamps

00:00:00 — Introductions
00:02:49 — The last month in AI news
00:10:19 — What is Grafana Assistant?
00:12:30 — What does it mean for Assistant to use tools?
00:13:30 — What is MCP?
00:16:56 — What is a CLI in an AI context?
00:19:17 — What are skills?
00:24:37 — Can you use a custom MCP server with Grafana Assistant?
00:25:55 — Why team MCP?
00:28:22 — The context-window cost of MCP tool definitions
00:31:50 — Tool discoverability
00:39:25 — What is gcx, and why a CLI?
00:46:13 — Demo: creating an SLO with MCP vs gcx
00:52:58 — Benchmarking MCP vs gcx with o11y-bench
00:58:10 — How MCP and gcx relate to Grafana Assistant
01:02:35 — k6 MCP vs Grafana MCP
01:03:00 — When to use which (decision guide)
01:04:29 — Adding the OSS Grafana MCP to Assistant
01:05:41 — Turtles all the way down

Resources

MCP and CLI for agents

My Context Horizon piece: Unlearnings from building Grafana Assistant

Mcp on Nicole van der Hoeven

When to use Grafana Assistant vs MCP vs GCX: Grafana MCP

Transcript

Resources

AI Week Day 2: Building (Grafana AI Week — Tuesday)

Transcript

CLI and MCP Basics for AI Agents (The Context Window #04)

Timestamps

Resources

MCP and CLI for agents

Related reading

More from The Context Window