# Connect ChatGPT, Claude and Cursor to one database

ADoor is your data account, on real Postgres. ChatGPT, Claude, Grok and Cursor connect to the same tables over MCP, each with its own scoped access, and it's free to start.

## How to connect ChatGPT to a database

1. Sign in to ADoor.
2. In ChatGPT, open Plugins, then Add custom MCP server, with the URL `https://app.adoor.halmoslabs.com/mcp`. Choose OAuth.
3. ADoor opens in your browser. Choose the space ChatGPT may reach, and approve.
4. Ask ChatGPT to set up a table, or to query one.

**You:** Start a list of the apartments I'm looking at: address, rent, size, and whether I've visited.

**ADoor:** Created Apartments in your space and added 2 entries

**Apartments**

| Address | Rent | Size | Visited |
| --- | --- | --- | --- |
| 14 Elm St | 1,850 | 52 m² | Yes |
| 3 Harbor Rd | 2,100 | 61 m² | No |

## Connect Claude, Grok and Cursor over MCP

The same MCP URL works in each one, with the same OAuth sign-in. In Claude: Customize, then Connectors, then add a custom connector. In Grok: Connectors, then New Connector, then Custom. In Cursor: add ADoor under `mcpServers` in `~/.cursor/mcp.json`. The [connect guide](/docs/connect/) has the steps.

Connect them to the same space, and the next day Claude can answer “Which apartments haven't I visited yet?” from the table ChatGPT started.

## What you get

- **Tables, with fields and rows.** Describe what you want to keep track of, and the assistant sets it up in your space.
- **SQL.** Standard Postgres SQL through MCP and the HTTP API, within the access each caller has.
- **Scoped access.** Each assistant, app and person gets its own access: a space, one table, some fields or certain entries. Read, add, edit and delete are separate.
- **Approvals.** An assistant that needs more access asks, and you approve or deny.
- **History.** Every change records who made it and the before and after.
- **Export.** CSV or JSON, whenever you want.

## Why not a raw Postgres MCP server?

A Postgres MCP server you run yourself on [Neon](/compare/adoor-vs-neon/) or [Supabase](/compare/adoor-vs-supabase/) is better when you already own the database and want full control of roles. It usually means a connection string, and often one with full write access, handed to each assistant.

ADoor gives out no connection string. Each assistant signs in, gets only the access you approve, and you revoke it in one place. Memory tools such as [Mem0](/compare/adoor-vs-mem0/) and [Zep](/compare/adoor-vs-zep/) are better at pulling facts out of conversations. ADoor keeps plain tables people can read and correct.

## Questions people ask

### How do I connect ChatGPT to a database?

Add ADoor as a custom MCP server in ChatGPT, at https://app.adoor.halmoslabs.com/mcp, and sign in with OAuth. Choose the space ChatGPT may reach, and it can set up and query tables there.

### Can Claude access databases?

Yes, through an MCP connector. Add ADoor as a custom connector in Claude, and it reads and writes the tables in the space you choose, with the access you approve.

### Is it real Postgres?

Yes. Tables are ordinary Postgres data, and assistants and scripts run standard Postgres SQL through MCP and the HTTP API, within their access.

### What does it cost?

A free workspace holds 1 space with 3 tables and 100 rows. Pro is $10 a month per workspace, with 1 GB of storage and no per-GB charges. See [pricing](/pricing/).

## Start

[Get your account](https://app.adoor.halmoslabs.com/signup)

## Related

- [How to connect ChatGPT and Claude to the same data](/use/connect-chatgpt-and-claude/)
- [How to give an AI agent safe, scoped access to data](/use/scoped-access-for-agents/)
- [MCP reference](/docs/reference/mcp/)
- [Pricing](/pricing/)

Updated October 20, 2026.