Gatlas in Claude, ChatGPT and Cursor
Ask about any of your companies in plain language, from your own AI agent, such as Claude, ChatGPT or Cursor. It reaches Gatlas through MCP, the open standard that AI agents such as Claude and ChatGPT use to connect to other software.
Example conversation
Agent capabilities
A dataset is a table Gatlas builds, written up with a description and column notes. A brief is prose, such as a company overview, kept current from a company’s files and datasets.
Your agent connects with an agent key. With a key that can write, it can also build datasets and artifacts (dashboards, documents and decks), and send setup links after you confirm.
Answer trace
Orders by week
Each company’s orders this week and the week before, by sales channel.
- company
- The company that took the orders
- channel
- Online, Retail or Wholesale
- this_week
- Orders taken this week
- week_before
- Orders taken the week before
| company | channel | this_week | week_before |
|---|---|---|---|
| Alder Goods | Online | 412 | 468 |
| Cedar Labs | Wholesale | 1,086 | 1,154 |
| Rowan Supply | Retail | 233 | 241 |
- data_as_of
- 2026-11-20T01:10:00.000Z
Safeguards
- Access
- Each person and each agent sees only the companies you grant
- Missing values
- A missing value shows as missing
- Partial results
- Marked as one page of several, with a warning against totaling them
- Changes
- Every action that changes data is marked as one, so your agent can ask you first
- Read-only keys
- Read data and send feedback, and change nothing else in Gatlas
- Credentials
- Given by whoever connects the tool, you or the person a setup link went to, and used only for that tool
- Request log
- Every request an agent makes is logged, with a summary of what it asked and its outcome
Put your data to work
Create your workspace and connect each company’s tools, or send a setup link to whoever holds the login. Your own AI agent can make the links and build the datasets.