Introducing Datachef for Claude

1:38

Table of Contents

• What is the Datachef connector for Claude?
• Why not just connect Claude to each app?
• How does the connector work?
• What can you ask it?
• How do you set it up?

What is the Datachef connector for Claude?

The Datachef connector for Claude makes it easier than ever to manage your analytical data sets and get answers from your business data.

You can manage your sync settings, build and share data models, and create visualizations that help you tell stories with data - all from Claude.

Why not just connect Claude to each app?

You can. Claude can connect to Salesforce, HubSpot, Postgres and plenty of other systems on their own, and for a quick lookup in one system ("what's the next step on the Acme opportunity?") that works fine.

It falls apart when the question spans more than one system, which most good questions do. A few reasons why:

1. No joins across systems - say you want to know which Marketo programs produce customers who actually use the product. That's Marketo leads, Salesforce opportunities and usage events from your product database in Postgres. Connected separately, Claude has to pull rows out of each one and try to stitch them together in the conversation. In Datachef, all three are tables in one place, so it's one SQL query.
2. You'd be building a data stack - to answer properly, the agent needs your data somewhere it can query. That means storage, ETL to pull data out of every source and keep it fresh as schemas change, and models on top. An agent can try to build all of that for you, but when it's done, you own it. And if you want your colleagues to use it too, you have to deploy it and keep it running. Why do all that when you can just use Datachef?
3. No shared definitions - every company has its own definition of things like a qualified opportunity, an active customer or net revenue. That definition doesn't live in any of your source systems, so an agent invents one each time it's asked, and won't tell you which one it chose. Two people (or two agents) ask the same question and get two different numbers. What you need is one place where those definitions live, what Andreessen Horowitz calls a system of intelligence that sits on top of your systems of record. In Datachef, your definitions are saved SQL models, so Claude uses your definition, the same one your dashboards and your colleagues use.
4. Your production database - pointing an AI agent at the Postgres or MySQL database behind your product means its queries compete with your customers' traffic (same reason you don't run analysis against production). Datachef queries a synced copy instead.

How does the connector work?

Datachef is a full-stack data platform: it syncs your sources on a schedule, stores them as tables, and keeps your business logic as versioned SQL models. The connector gives Claude a set of tools to work with all of that.

Claude can:

1. Explore - list your tables (grouped by the source they came from), see columns and your team's notes on them, and read your saved models
2. Query - run read-only SQL against your tables and models, one SELECT at a time
3. Check your syncs - see what each connection syncs, how often, and whether the last sync worked
4. Trace lineage - see what a model is built on and what depends on it
5. Make changes, if you let it - add a column to a sync, kick off a sync, or save a query as a new model

And a few things it can't do:

• It never sees your connection credentials. Those stay in Datachef.
• It can't write to your data. Queries are read-only and run through the same checks as Datachef's own SQL editor.
• It only sees the workspaces you choose when you connect.
• It can't change anything unless you tick Allow changes, and even then, only what you could change yourself. Every query and every model it saves is marked "via Claude" in Datachef, so you can see what it did.

What can you ask it?

Here are some examples of questions and commands you can issue to Datachef from Claude:

• "Which marketing programs sourced or influenced the pipeline we closed this quarter, and what's the average deal size for each?"
• "How does the average deal size of marketing attributed pipeline compare to uninfluenced pipeline?"
• "Of the customers who signed in the last six months, which ones haven't logged in since onboarding?"
• "Is the Salesforce connection up to date? Did any sync fail this week?"
• "what is our open rate of emails by industry over the past 6 months?"

How do you set it up?

1. In Claude, open Settings → Connectors and add Datachef
2. Sign in to Datachef
3. Select the workspaces Claude can use (and Allow changes, if you want it to)
4. Choose Allow access, and you're back in Claude, connected

It works in Claude on the web, desktop and mobile, and in Claude Code. If you don't have data connected yet, you can load the sample dataset in Datachef with one click and try it on that.

Want to see it on your own data? Contact us to get a demo, or try Datachef free.