> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nodi3d.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Coding Agents: Model in Nodi with Claude Code, Codex, Gemini CLI and Cursor

> Connect Claude Code, Codex, Gemini CLI or Cursor to Nodi for desktop and build or edit node graphs by prompting. One command, no MCP server configuration.

Nodi for desktop embeds a local control server that coding agents drive through the `@nodi3d/agent` CLI. Ask your agent to model something and every change appears live on the Nodi canvas.

> [!NOTE]
>
> This integration requires [Nodi for desktop](https://docs.nodi3d.com/nodi-desktop/overview.md). The web editor cannot connect to a coding agent; its Agent panel shows the same setup guide as this page.

## Setup

**1. Install Nodi for desktop**

Download the app from [nodi3d.com/desktop](https://nodi3d.com/desktop) and launch it.

**2. Install the Nodi instructions for your agent**

Run this in your terminal. It detects the agents installed on your machine and installs the Nodi instructions for each of them:

```bash theme={null}
npx @nodi3d/agent@latest setup
```

| Agent | What gets installed |
| - | - |
| Claude Code | The `nodi` skill under `~/.claude/skills/` |
| Codex | A block appended to `~/.codex/AGENTS.md` |
| Gemini CLI | A block appended to `~/.gemini/GEMINI.md` |
| Cursor | A project rule at `.cursor/rules/nodi.mdc` (run setup inside your project) |

Re-run `setup` after upgrading the package so the installed skill stays current.

**3. Connect and model**

With Nodi for desktop running, start your agent and invoke the skill. In Claude Code:

```text theme={null}
> /nodi Generate gradient gyroid infill in sphere
```

The agent connects to the open project, adds and wires nodes, evaluates the graph and takes viewport screenshots to check its work. Reference images work too: `/nodi replicate <image path>` reproduces a part from photos or drawings.

The full guide is at [nodi3d.com/agents](https://nodi3d.com/agents/); **Help → AI Agent Guide** in the app opens it in your browser. The blog post [AI-Assisted Modeling in Nodi](https://nodi3d.com/blog/ai-modeling) walks through modeling from photos, generating Expressions and sending the result into analysis.

## The Agent panel

The **Agent** tab in Nodi for desktop shows what the connected agent is doing:

* **Connection**: the window the agent is connected to. With several windows open, click **Use for agent** in the window you want it to drive, or tell the agent the window number shown as `Nodi [n]` in the title bar.
* **Activity log**: every agent command and its result in real time.
* A status chip in the top-right corner of the canvas shows the agent's current state while it works.

When the agent mentions a node in its reply, the link pans the editor to that node and selects it.

## How it works

The control server listens on localhost only and starts with the app. Agent edits go through the same graph engine as the UI, so the canvas and viewport update live and Undo works as usual. Nothing leaves your machine.

## Troubleshooting

* **The agent cannot launch the app**: start Nodi for desktop manually, or install it from [nodi3d.com/desktop](https://nodi3d.com/desktop).
* **Commands fail with a list of windows**: more than one window is open and none is selected. Click **Use for agent** in the window you want, or run `nodi-agent use <n>`.
* **The installed skill is out of date**: run `npx @nodi3d/agent@latest setup` again.


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