# Connect datatomcp in your AI app

`https://datatomcp.com/install.md` · for AI agents helping a person connect datatomcp

datatomcp is a team's data, ready for every AI agent: PostgreSQL, MySQL, Stripe, HubSpot, Airtable, Google Sheets, Excel and CSV files in one workspace, with approved definitions, linked datasets and an audit of every question. You connect to it as a **remote MCP server**. There is nothing to clone, build or run locally.

Read this as guidance, not as a script. Follow only the path for the app you are running in, and show the person only that path, in their language.

## Start here

1. **Check for datatomcp tools already in this conversation** (`list_datasets`, `workspace_overview`). If they are there, skip to **Verify the connection**. Don't add a second connection.
2. **Work out which app you are running in** from reliable session information. "Claude" alone doesn't tell Claude chat apart from Claude Code. If you can't tell, ask one question: "Which app are you using?"
3. **Prefer the app's own connector directory or remote MCP settings.** If you can run the setup yourself and are allowed to, do it, and let the person complete the sign-in. If you can't change the app's settings (a chat app can't), give the person the matching steps below and one clear next action, then wait.
4. **If an organisation policy blocks custom connectors**, say that an administrator has to allow datatomcp. Don't suggest another app or a workaround.

## Connection details

| Setting | Value |
| --- | --- |
| Name | datatomcp |
| Remote MCP server | `https://app.datatomcp.com/mcp` |
| Transport | Streamable HTTP |
| Authentication | OAuth in the browser (PKCE). The client finds the sign-in from the URL |
| Account | Signing in creates a free account if the person has none (no card) |

Use the exact URL, including `/mcp`. datatomcp never needs an API key, a client secret, a pasted token or a static `Authorization` header in an MCP client; the app handles the sign-in.

## Claude (claude.ai, Claude Desktop, mobile)

1. Open [datatomcp in the Claude Directory](https://claude.ai/directory/datatomcp), or in Claude open **Customize → Connectors** and search "datatomcp".
2. Click **Connect**, sign in to datatomcp and allow access.
3. Ask about the data in any chat. Desktop and mobile use the same connector.

On a Team or Enterprise plan an owner adds the connector first under **Organization settings → Connectors**. If the directory isn't available, **+ Add → Add custom connector** with the URL above works too. Don't edit `claude_desktop_config.json` or run `claude mcp add` for the chat apps.

## Claude Code

Run in the terminal (the person's terminal, not a sandbox that doesn't configure their app):

```sh
claude mcp add --transport http datatomcp https://app.datatomcp.com/mcp
```

Then inside Claude Code run `/mcp`, pick **datatomcp** and authenticate in the browser.

## Cursor

Open this link to install (Cursor asks to confirm, then to log in; that's the datatomcp sign-in):

`cursor://anysphere.cursor-deeplink/mcp/install?name=datatomcp&config=eyJ1cmwiOiJodHRwczovL2FwcC5kYXRhdG9tY3AuY29tL21jcCJ9`

Or add to `~/.cursor/mcp.json`:

```json
{ "mcpServers": { "datatomcp": { "url": "https://app.datatomcp.com/mcp" } } }
```

## Codex

```sh
codex mcp add datatomcp --url https://app.datatomcp.com/mcp
codex mcp login datatomcp
```

The second command opens the browser to sign in.

## ChatGPT

In ChatGPT open **Settings → Plugins** (or **Apps**) and search "datatomcp". If it isn't listed for the account yet, use **+ → Add custom MCP server** with the URL above and OAuth; some accounts need developer mode first (**Settings → Apps → Advanced settings**), and on a workspace an admin may need to allow it.

## Other apps

| App | How |
| --- | --- |
| VS Code | Command Palette → **MCP: Add Server** → HTTP → the URL, name it `datatomcp`, sign in when asked |
| Gemini CLI | `gemini mcp add -s user --transport http datatomcp https://app.datatomcp.com/mcp`, then `/mcp auth datatomcp` |
| OpenCode | in `opencode.json`: `{ "mcp": { "datatomcp": { "type": "remote", "url": "https://app.datatomcp.com/mcp" } } }`, then `opencode mcp auth datatomcp` |
| Antigravity, Windsurf | MCP config: `{ "mcpServers": { "datatomcp": { "serverUrl": "https://app.datatomcp.com/mcp" } } }`, then authenticate |
| Cline | `{ "mcpServers": { "datatomcp": { "type": "streamableHttp", "url": "https://app.datatomcp.com/mcp" } } }` |
| Local-only (stdio) clients | `npx -y mcp-remote https://app.datatomcp.com/mcp` as the command; it handles the sign-in |

## Verify the connection

1. Call `workspace_overview` (or `list_datasets`). A new account has an example dataset, `sample_sales`, marked as example data.
2. Tell the person it works, then offer one first question, for example: "What was total revenue by region in sample_sales?" Answer it with `aggregate` and share `meta.view_url` so they can check the number.
3. To use their own data, they add a file, a Google Sheet or a database at https://app.datatomcp.com/app/new. To look around first without signing up: https://app.datatomcp.com/demo.

## What the agent gets

Read-only data tools: `workspace_overview`, `list_datasets`, `describe_dataset`, `distinct_values`, `query_rows`, `aggregate`. Also `chart` (draws the answer in the chat), `create_dashboard`, `list_dashboards`, `refresh_dashboard` (new pages in the workspace; data is never changed) and `report_issue`. Totals are computed by a database, every answer names the dataset version it came from, and the team's approved definitions are used by name.

Agents also help build the workspace, and a person approves: `propose_definition`, `propose_link`, `propose_term` and `propose_brief` suggest a metric definition, a link between datasets, a glossary term or the business brief, and each waits in the app until an owner or editor approves it. `connect_source` starts a connection to a database, app or file, and `activity` shows what agents asked and changed. For a business number, the agent first uses the team's approved definition and checks that it ties; when none exists, it proposes one. An owner can set any agent to Analyze to keep it read-only.

Questions or problems: hello@datatomcp.com · Docs: https://datatomcp.com/guides/add-remote-mcp-server · Security: https://datatomcp.com/security