Disclosure: This post is sponsored by Sigma. As a Sigma-recognized brand advocate, I’ve been compensated for this content. As always, all opinions, bad analogies, and questionable pop culture references are entirely my own.
Sigma dropped something yesterday and I stayed up late playing around with it
Not in a “cool feature, moving on” kind of way. More in a “okay wait, let me actually think through what this means” kind of way. Because on the surface, the Sigma MCP Server sounds like a tidy little integration announcement. Connect your AI assistant to Sigma, ask questions, get answers. Yes please
But the more I sat with it, the more I realized this is a different kind of move. Not “we made Sigma more accessible.” More like “we brought your governed data to wherever you already live.” And I think that distinction matters more than the blog makes it sound.
Let me explain what I mean.
What Even Is MCP?
Fair question. MCP stands for Model Context Protocol — it’s essentially a standardized way for AI assistants like Claude or ChatGPT to connect to external tools and data sources. Think of it like a universal translator between your AI chat interface and the systems that actually hold your data.
Sigma built an MCP Server that plugs directly into that protocol. The result: your AI assistant can now search, explore, and query your Sigma organization — your actual governed workbooks, data models, and warehouse connections — without you having to leave the conversation.
No more tab-switching. No more “let me go pull that.” Just a question, and an answer, grounded in real data.
The Three Things It Can Actually Do
Sigma’s MCP Server is built around three capabilities, and I want to be real with you about what each one means in practice.
Search is exactly what it sounds like — you can ask your AI assistant to find stuff in your Sigma org. “Do we have anything on customer retention?” and Sigma’s semantic search surfaces the relevant workbooks, data models, and connections. This alone is genuinely useful if you’ve ever worked in an org where nobody can find anything because someone named a workbook “Final_FINAL_v3_USE THIS ONE.”
Analyze is where it gets fun. Once the AI finds the right asset, it can run queries and return results right there in the conversation. Ask a follow-up. Drill into an anomaly. Cross-reference another model. You’re building on a thread, not starting over with every question. This is the part that made me do the TV-pointing thing.
Build is coming in May (at time of writing), and it’s going to let the AI generate full Sigma workbooks and dashboards from a prompt. The whole workflow — from “I have a business question” to “here’s a finished dashboard” — without leaving the chat. I’ll definitely be writing about that one when it drops.
“But Is My Data Safe Though?”
I knew this was coming, and honestly, it’s the right question to ask. Here’s the part that actually impressed me when I dug into the docs.
The Sigma MCP Server doesn’t create a new security surface. It inherits your existing Sigma permissions — all of them. Account-level access, connection-level access, column-level security, row-level security, workspace permissions. If you can’t see a column in Sigma, the AI assistant can’t see it either. If a workbook isn’t shared with you, it’s invisible to the assistant. The AI sees exactly what you see — nothing more, nothing less.
That’s not a small thing. One of the real concerns with AI and enterprise data is over-permissioning — the AI getting access to things the user shouldn’t have. Sigma sidesteps that entirely by just… using your credentials!
What This Looks Like in Real Life
Let me give you a few scenarios that I think actually resonate.
The Monday morning question. Your VP pings you: “Hey, what was our profit margin by product family last quarter?” Normally, that’s a 15-minute detour. With the MCP Server connected, that’s a 30-second conversation with Claude. Real answer, governed data, no context-switching.
The anomaly nobody caught. Here’s the one that kept me up last night (in a good way). You can set up an agent that monitors your key Sigma metrics on a schedule and automatically investigates when something looks off. Revenue down week-over-week? The agent pulls the relevant workbooks, slices by region and channel, and drops a plain-English summary in your Slack before your morning standup. That’s not a feature — that’s a whole new way of operating.
The “where’s the data?” problem. Every org has this. Someone new joins, asks “do we have data on X?”, and the answer is usually “…probably? Try asking [person who’s been here forever].” With the MCP Server, that question goes to the AI instead, which searches your entire Sigma org and surfaces what exists. Knowledge transfer, solved.
Setting It Up (It’s Not Scary, I Promise)
Getting connected takes about five minutes if you already have a Sigma account. You grab your MCP URL from your Sigma profile (Profile > MCP > Sigma MCP), then in Claude you go to Customize > Connectors > Add custom connector, drop in the URL, and authenticate with your Sigma login. That’s it. OAuth handles the rest.
The Bigger Idea Here
The data tools conversation for the last decade has been about making BI more accessible — better UI, drag-and-drop, natural language queries inside the tool. All good progress. But the Sigma MCP Server is a different move. It’s not making Sigma more accessible. It’s making Sigma’s data accessible from wherever you already are.
That’s a subtle but meaningful shift. People aren’t going to stop living in Slack and email and AI chat interfaces. The question is whether your data can meet them there. Sigma’s answer is yes — with governance intact, with permissions enforced, with your existing data models doing the work.
I have a lot of opinions about where this goes from here (the Build capability in particular is going to be a moment), and I’ll keep writing about it. But for right now, if you’re a Sigma customer and you haven’t connected the MCP Server to your AI assistant yet, go do it today. Seriously. It’s five minutes and it’ll make you do the TV-pointing thing.
Go Forth and Viz.
Questions? Tried this yourself? Find me on LinkedIn. I want to know what you’re building with it.