Skip to main content

​
Prerequisites

  • A TeamDuo account and a team
  • A PostgreSQL, MongoDB, or BigQuery database you can administer
  • An MCP client — Claude, ChatGPT, or anything that speaks remote MCP
1

Create a team

Teams own datasources and decide who can query them. Create one, then invite the people who should have access.
2

Connect a database

Add a datasource and follow the wizard. It generates the read-only credential for you to run — TeamDuo never creates it on your behalf.If your database sits behind a firewall, the wizard also gives you TeamDuo’s egress address to allowlist. See Network access.
3

Choose what to expose

Nothing is readable until you say so. Select the tables and columns agents may see.See Connect a datasource for the full walkthrough.
4

Document what the schema doesn't say

Generate documentation, review the drafts, and publish the ones that are right. This is the step that decides whether an agent writes the query you meant.See Documentation.
5

Connect your agent

Add https://api.teamduo.ai/mcp as a custom connector and approve it. See Connect an agent.
6

Ask a question

A typical first exchange:

​
What a refusal looks like

Queries touching anything outside the exposed surface are refused by name — “table public.salaries is not exposed” rather than a generic error. An agent that is told what was blocked corrects itself; one given a vague error retries blind.
Start with sample_rows on an unfamiliar table. It shows the shape of the data without spending a query on an exploratory SELECT *.