# Setup

Get an AI agent building dcupl projects in three steps: install the CLI, install the dcupl skill in Claude Code, then confirm the agent can drive both. The whole setup takes a couple of minutes.

## Step 1: Install the CLI

Install the CLI globally so the `dcupl` binary is available everywhere, then verify the resolved engine versions.

```bash
# install the CLI globally
npm install -g @dcupl/cli

# verify + check resolved @dcupl/* versions
dcupl version
```

If it prints versions, you're ready. If you hit `command not found`, see [CLI Installation](/cli/installation).

## Step 2: Install the dcupl skill

The dcupl skill is a Claude Code plugin. Add the marketplace and install it from within Claude Code:

```text
/plugin marketplace add dcupl/dcupl-plugins
/plugin install dcupl@dcupl-plugins
```

Once installed, the agent has the dcupl domain knowledge it needs to use the CLI correctly — no extra prompting required. The plugin works in Claude Code and in the Claude desktop app. Every change to the plugin is an update: pull it with `/plugin marketplace update dcupl-plugins`.

## Step 3: Verify the agent can drive it

Confirm everything is wired up by asking the agent to do something concrete:

> **You:** Scaffold a new dcupl project and show me the structure.

The agent should reach for the skill, run `dcupl init` with a `--starter` (its shell can't answer the interactive starter prompt), and walk you through the generated files. If it scaffolds a project and explains the layout, your setup works — you're ready to build.

> **Note:** **Local-only by default.** Scaffolding and querying need no account. To let the agent *deploy or sync to the cloud*, create a project in the [dcupl Console](https://console.dcupl.com) and have the agent run `dcupl config set` with the project ID and API key (see [Cloud Sync](/cli/cloud-sync)).

> **Tip:** Every command accepts `--json` for machine-readable output. The skill already knows this, but it's the single most useful flag if you wire dcupl into your own agent — see [Why it works](/agentic/why-it-works).

## Docs for agents

The skill carries the command patterns; for anything else, point the agent at these docs in a form it reads directly:

- **Every page as markdown.** Append `.md` to any docs URL, for example [/sdk/quick-start.md](/sdk/quick-start.md).
- **An index.** [/llms.txt](/llms.txt) lists every page with its description and links the full-text files.
- **Full text per section.** One file per section, each sized to fit one agent context: `llms-getting-started.txt`, `llms-sdk.txt`, `llms-sdk-api.txt`, `llms-sdk-advanced.txt`, `llms-sdk-guides.txt`, `llms-cli.txt`, `llms-agentic.txt`, `llms-console.txt`, `llms-learn.txt` and `llms-playground.txt`; [/llms-full.txt](/llms-full.txt) holds all of them in one file.
- **A deployed API.** A REST API (RAPI) deployed from the Console exposes an MCP endpoint where it is turned on, so an agent can query the data without the CLI: see [MCP endpoint](/console/rapi/mcp).

## Next steps

- [Recipes](/agentic/recipes) — Prompt-to-action walkthroughs you can try now
- [The dcupl skill](/agentic/the-skill) — What the skill teaches the agent
- [Quickstart](/cli) — Build your first project end to end
- [Why it works](/agentic/why-it-works) — The design behind agent-readiness