> ## Documentation Index
> Fetch the complete documentation index at: https://docs.doglytdc.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Install with one command

> One command (or a prompt pasted into the AI) installs the MCP connection, the Skill, and per-function .md files in every AI on your computer.

The `doglytdc-scripts` command automatically does what the [Configure manually](https://docs.doglytdc.dev/en/scripts/mcp/connect.md) page explains: it creates a connection key for each AI, writes the MCP configuration in the right place, and installs the Skill with one `.md` file per tool group. You do not copy or paste any keys.

## Over the internet: chat apps (Claude.ai, Claude Desktop, ChatGPT)

> [!WARNING]
>
> The command **does not work inside Claude.ai or ChatGPT**. Those apps run commands on an isolated machine with no internet (npm answers `ENOTFOUND` / `getaddrinfo`), and nothing installed there reaches the app. Use the connector below.

Chat apps connect through the MCP address and ask you to sign in to DoglyTdc — no key and no terminal.

1. Copy the connector address from **MCP connections** on the site (or use `https://api.doglytdc.dev/v1/scripts/mcp`).
2. **Claude.ai / Claude Desktop:** Settings → Connectors → **Add custom connector** → paste the address → **Connect**.
3. **ChatGPT:** Settings → Apps & connectors → Advanced → turn on **Developer mode**; then **Create**, paste the address and choose OAuth.
4. Sign in to DoglyTdc and click **Authorize**. The app shows up in MCP connections, with identity, approval rules and **Revoke**.

Agents with a terminal and internet on your computer (Claude Code, Codex, Cursor, Gemini CLI, VS Code, Windsurf) use the command below.

## Paste into your AI agent

```text theme={null}
Instale o DoglyTdc Scripts e conecte a este agente usando `npx -y doglytdc-scripts@latest setup`. Primeiro rode com --dry-run e me mostre o plano de instalação para eu aprovar; depois rode de novo com --yes e confira se funcionou. Importante: só rode isso se você tiver um terminal com internet no MEU computador. Se você for um app de chat (Claude.ai, ChatGPT, Claude Desktop) ou rodar comandos numa máquina isolada sem internet, não rode nada: me explique como adicionar um conector personalizado com o endereço https://api.doglytdc.dev/v1/scripts/mcp (Configurações → Conectores → Adicionar conector personalizado → colar o endereço → Conectar → entrar no DoglyTdc e aprovar).
```

![Screenshot: Approving the install command](https://mintcdn.com/dogly-tdc/qOnot79kPtISSKbW/images/scripts/scripts-cli.png?fit=max\&auto=format\&n=qOnot79kPtISSKbW\&q=85\&s=0d89568bfd1ccf129ac3fb23b20fdbe7)

Approving the install command

## Or run in the terminal

> [tab] Node (any system)
>
> ```bash theme={null}
> npx -y doglytdc-scripts@latest setup
> ```

> [tab] Windows (PowerShell)
>
> ```powershell theme={null}
> irm https://ucsyxzpdbnuyehizezvb.supabase.co/functions/v1/doglytdc-scripts-api/scripts/install.ps1 | iex
> ```

> [tab] macOS / Linux
>
> ```bash theme={null}
> curl -fsSL https://ucsyxzpdbnuyehizezvb.supabase.co/functions/v1/doglytdc-scripts-api/scripts/install.sh | sh
> ```

> [!NOTE]
>
> Requires Node.js 18 or newer. No Node.js? Use [Configure manually](https://docs.doglytdc.dev/en/scripts/mcp/connect.md).

## What happens

**1. Plan**

The command looks for installed AIs and shows exactly which files it will create or change. Nothing changes before your approval.

**2. Approval**

In the terminal, answer `s`. Through the AI: it shows the plan in the chat and, once you approve, runs again with `--yes`.

**3. Code in the browser**

Opens the DoglyTdc **Conectar IA** (Connect AI) page with an 8-letter code (e.g., `BHKT-QXZM`). Check that it matches the terminal, choose what the AIs can do, and tap **Aprovar** (Approve). The code is valid for 10 minutes and can only be used once.

**4. Installation**

Each AI receives its own key (shown in **Scripts › Conexões MCP** (Scripts › MCP Connections) with the AI and computer names), the MCP connection, and the Skill.

**5. Verification**

The command calls `whoami` with each AI's key and displays your account. Restart the AIs to load MCP.

## Where everything goes

| AI | MCP connection | Skill and per-function .md files |
| - | - | - |
| Claude Code | `claude mcp add-json --scope user` (or `~/.claude.json`) | `~/.claude/skills/doglytdc-scripts/` |
| Codex | `~/.codex/config.toml` | `~/.codex/skills/doglytdc-scripts/` + notice in `~/.codex/AGENTS.md` |
| Cursor | `~/.cursor/mcp.json` | Project's `.cursor/rules/doglytdc-scripts*.mdc` |
| Gemini CLI | `~/.gemini/extensions/doglytdc-scripts/` extension | `GEMINI.md` + `reference/` in the same extension |
| VS Code / Copilot | User's `mcp.json` | Project's `.github/instructions/doglytdc-scripts.instructions.md` |
| Claude Desktop | `claude_desktop_config.json` (`mcp-remote` bridge) | `~/.doglytdc/scripts/doglytdc-scripts-skill.zip` to upload in Skills |
| Windsurf | `~/.codeium/windsurf/mcp_config.json` | `~/.doglytdc/scripts/skill/` |

Cursor and VS Code rules are per project: they are created when you run inside a folder containing `.git`, `package.json`, or `doglytdc.json`, or with `--project`. A copy of the Skill is always kept in `~/.doglytdc/scripts/skill/`.

## Commands

| Command | What it does |
| - | - |
| `setup` | Plan → approval → login → installation → verification (default) |
| `login` / `logout` | Creates / revokes this computer's keys |
| `status` (or `doctor`) | What is installed in each AI and whether it connects |
| `update` | Downloads the latest Skill and .md files |
| `skills [nome]` | Lists the Skill or shows a file (e.g., `skills codigo`) |
| `uninstall` | Removes only what the command installed and revokes the keys |

## Options

| Option | Purpose |
| - | - |
| `--agent a,b` | Only these AIs: `claude-code`, `codex`, `cursor`, `gemini`, `vscode`, `claude-desktop`, `windsurf` |
| `--project` | Skill and rules in this folder (the MCP connection remains under your user account) |
| `--yes`, `-y` | Already approved (the AI uses it after you approve the plan in the chat) |
| `--dry-run` | Only shows the plan |
| `--no-browser` | Does not open the browser; the link appears in the terminal |

Example:

```bash theme={null}
npx -y doglytdc-scripts@latest setup --agent codex,cursor --project --dry-run
```

## Security

* The key never appears in the AI chat or terminal: it stays in `~/.doglytdc/scripts/credentials.json`, with permissions restricted to your user account.
* Codex and Claude Desktop receive the key through a variable (`DOGLYTDC_AUTH`), not on the command line.
* Before changing a configuration file, the command makes a `.bak` copy and only modifies the `doglytdc-scripts` entry.
* Never approve a code someone else sent you: whoever holds the approved code receives the keys.
* Revoking a connection in **Conexões MCP** immediately disconnects that AI.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.