KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #ai-agents #dsh-plugin #codex #developer-tools #claude-code #windows #desktop-app #typescript #rust #ai-agent #cli

google-labs-claude

★ 1 Open GitHub ↗

Claude Code plugin and MCP server that connects Google Labs creative tools to Claude: Veo video generation, Stitch UI design, and Pomelli brand content — all three generating against one shared brand identity.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Google Labs → Claude Code & Claude Desktop

**Generate video with Veo, design interfaces with Stitch, and create brand content with Pomelli — all from one project, against one brand identity.**

[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://raw.githubusercontent.com/holm-digital-io/google-labs-claude/main/LICENSE)
[![Claude Code Plugin](https://img.shields.io/badge/Claude%20Code-plugin-8A63D2)](https://docs.claude.com/en/docs/claude-code/plugins)
[![MCP](https://img.shields.io/badge/MCP-server-0A7EA4)](https://modelcontextprotocol.io)
[![Node](https://img.shields.io/badge/node-%E2%89%A518-339933)](https://nodejs.org)

**google-labs-claude** is a Claude Code plugin and MCP server that connects three Google Labs creative tools — **Pomelli** (brand identity and ad content), **Stitch** (UI design), and **Flow / Veo / Omni** (video generation) — to Claude Code and Claude Desktop. What sets it apart: all three tools generate against a single shared brand identity stored in a `brand/BRAND.md` file. Define your brand once, and your ad creative, interface design, and promo video all come out in the same colors, typography, and tone.

Installation is one line and requires no npm download — the MCP server is embedded in the plugin as a single file. Node.js 18 or newer is the only prerequisite.

---

## Install in 60 seconds

```
/plugin marketplace add holm-digital-io/google-labs-claude
```

```
/plugin install google-labs
```

```
/google-labs-setup
```

Restart Claude Code, then run `/google-labs-setup`. The wizard finds missing API keys, shows you where to get them, and runs a live connection test.

---

## What can you build with it?

Ask in plain language — the right tool fires on its own.

| What you ask | Tool that fires | What you get |
|---|---|---|
| "Generate an 8-second product video from this image" | Veo (MCP) | MP4 in `assets/video/` |
| "Design a SaaS pricing page" | Stitch (MCP) | HTML/code in `assets/ui/` |
| "Extract brand DNA from my site and make 3 ads" | Pomelli (browser) | Images in `assets/brand/` + `BRAND.md` |
| "Build a launch kit: brand, landing page, and video" | All three | Full `assets/` tree, one brand identity |
| "Look at this screenshot and write a video prompt" | Gemini multimodal | Text output |

For the full automated pipeline, one command:

```
/google-labs-project Build a launch kit for https://example.com — brand identity, landing page, and an 8-second promo video
```

---

## Which tool does what?

| Tool | Purpose | Connection | Cost |
|---|---|---|---|
| **Pomelli** | Brand DNA, ad and campaign content | Skill (browser) | Free |
| **Stitch** | UI design, design systems | Official remote MCP | Free (API key required) |
| **Veo** | Video generation (text-to-video, image-to-video) | MCP in this repo | Billed per generation |
| **Nano Banana / Imagen** | Image generation and editing | MCP in this repo | Free tier, then billed |
| **Gemini (Omni-style)** | Reads images and video, produces plans and prompts | MCP in this repo | Free tier, then billed |
| **Flow web UI** | Scene editing, conversational Omni editing | Skill (browser) | Google One AI Pro |

**Selection rule:** when a job can be done through either the API or the browser, the plugin picks the API — it is faster, repeatable, and downloads files straight to disk. The browser is used only where no API exists: all of Pomelli, and Flow's scene-editing and Omni interface.

---

## How does the shared brand identity work?

Stitch and Veo cannot see your file system. So brand consistency is not achieved through a technical integration — the orchestrator skill writes the brand block into every prompt as explicit text.

```
<your project>/
├── brand/
│   └── BRAND.md        # colors, typography, tone, audience — every tool reads this
└── assets/
    ├── brand/          # Pomelli output
    ├── ui/             # Stitch output
    └── video/          # Veo / Flow output
```

The flow: **Pomelli** extracts brand DNA from your site → it is written into `BRAND.md` → **Stitch** designs the interface with that identity → **Veo** generates video with the same identity.

---

## Installation guide

### Claude Code (recommended)

1. Add the marketplace: `/plugin marketplace add holm-digital-io/google-labs-claude`
2. Install the plugin: `/plugin install google-labs`
3. Restart Claude Code
4. Run `/google-labs-setup` and follow the wizard

### Claude Desktop

1. Download `google-labs-mcp.mcpb` from the [Releases page](https://github.com/holm-digital-io/google-labs-claude/releases/latest)
2. Double-click the file
3. Claude Desktop asks for your API key in its own settings screen — no config file editing required

> The Desktop bundle includes the video, image, and multimodal tools. The browser-based Pomelli and Flow skills are exclusive to the Claude Code plugin.

### Other MCP clients (Cursor, VS Code, Gemini CLI)

```bash
claude mcp add google-labs -s user -e GEMINI_API_KEY=YOUR_KEY -- npx -y google-labs-mcp
```

Or add this to your client's config file:

```json
{
  "mcpServers": {
    "google-labs": {
      "command": "npx",
      "args": ["-y", "google-labs-mcp"],
      "env": { "GEMINI_API_KEY": "YOUR_KEY" }
    }
  }
}
```

---

## How do I get the API keys?

### 1. Gemini API key (required)

Powers the video, image, and multimodal analysis tools.

1. Open https://aistudio.google.com/apikey
2. Click **Create API key** — it is free
3. Add it to your shell profile:
   ```bash
   echo 'export GEMINI_API_KEY="your-key-here"' >> ~/.zshrc
   ```
4. Restart Claude

### 2. Stitch API key (for UI design)

1. Open https://console.cloud.google.com and select or create a project
2. Enable the **Stitch API**
3. Go to **APIs & Services → Credentials → Create credentials → API key**
4. Add it:
   ```bash
   echo 'export STITCH_API_KEY="your-key-here"' >> ~/.zshrc
   ```

Official documentation: https://stitch.withgoogle.com/docs/mcp/setup

### 3. Google One AI Pro (Flow web UI only)

Required only by the `google-flow-ui` skill. If you generate video through the Gemini API, you do not need a subscription.

### Optional settings

```bash
export GOOGLE_LABS_OUTPUT_DIR="$HOME/Downloads/google-labs"  # output folder
export VEO_MODEL="veo-3.1-generate-preview"                  # video model
export IMAGE_MODEL="gemini-2.5-flash-image"                  # image model
export OMNI_MODEL="gemini-flash-latest"                         # multimodal model
```

Model IDs change on Google's side. To see the IDs your account can actually reach, ask Claude to run the `list_models` tool with the `veo` filter.

---

## Tools, skills, and commands

### MCP tools (`google-labs` server)

| Tool | What it does |
|---|---|
| `setup_status` | Reports installation status and which capabilities are active |
| `list_models` | Lists the model IDs your API key can reach |
| `generate_video` | Starts a Veo video generation (from text or from an image) |
| `check_video_operation` | Polls the generation and downloads the finished video to disk |
| `generate_image` | Generates images or edits existing ones |
| `multimodal_generate` | Reads images and video, produces text, plans, or prompts |

### Skills

| Skill | When it fires |
|---|---|
| `google-labs-studio` | Multiple output types requested — the orchestrator |
| `google-pomelli` | Brand identity, ad and campaign content |
| `google-stitch` | Interface and screen design |
| `google-flow-ui` | Flow-specific scene editing and conversational Omni |

### Commands

| Command | What it does |
|---|---|
| `/google-labs-setup` | Setup wizard — checks keys, adds them, runs a live test |
| `/google-labs-project` | Starts the multi-tool production pipeline |

---

## Frequently asked questions

**Is this plugin free?**
The plugin itself is MIT-licensed and free. Pomelli and the Stitch API are also free. The only paid part is Veo video generation, billed per generation through the Gemini API. Image generation has a free tier.

**Do I need a Google One AI Pro subscription?**
No. A subscription is needed only if you want the Flow web interface (the `google-flow-ui` skill). Generating video through the Gemini API requires no subscription.

**Does Pomelli have an API?**
No. Pomelli has no official API, so the plugin drives it through Claude's browser tools. Login and CAPTCHA steps are left to you.

**How long does a Veo video take?**
Typically 1-3 minutes. `generate_video` returns an operation name immediately; you retrieve the result by calling `check_video_operation` a few times.

**Does it work in Claude Desktop?**
Yes. Download the `.mcpb` bundle from the Releases page and double-click it. The Desktop bundle includes the video, image, and multimodal tools; the browser-based Pomelli and Flow skills are Claude Code only.

**Stitch already has an official MCP server — why do I need this plugin?**
On its own, Stitch's server only does UI design. This plugin combines it with Pomelli's brand identity and Veo's video generation in a single project, so all three use the same colors, typography, and tone from `BRAND.md`.

**Do I need to install an npm package?**
No. The MCP server ships inside the plugin as a single embedded file, so nothing is downloaded from npm during installation. Node.js 18 or newer is enough. The npm package exists only for other MCP clients.

**Where does my data go?**
Your prompts and uploaded images go to Google's Gemini API. Browser-based skills use the Google Labs web apps directly. This plugin adds no intermediary server, and your API key stays on your own machine.

**Will Claude enter my password?**
No. The skills never enter passwords, verification codes, or CAPTCHAs — they stop and hand those steps to you. They also never click outward-facing buttons like "share" or "publish" unless you explicitly ask.

---

## Troubleshooting

| Symptom | Fix |
|---|---|
| Tools do not appear | Restart Claude — environment variables only reach newly started processes |
| `model not found` / 404 | Run the `list_models` tool, find an ID you can reach, and set `VEO_MODEL` or `IMAGE_MODEL` |
| Stitch will not connect | `STITCH_API_KEY` is missing, or the Stitch API is not enabled in your Google Cloud project |
| Video never becomes ready | Call `check_video_operation` a few more times; if it seems stuck, pass `raw: true` to see the raw state |
| "GEMINI_API_KEY is not set" | Add the key to your shell profile and restart Claude |

---

## Cost and safety notes

- **Veo video generations are billed.** The skills ask for confirmation before batch generation, but the bill is yours — see [Gemini API pricing](https://ai.google.dev/pricing).
- **The Flow web UI consumes your subscription credits.** When the same job can be done through the API, the plugin prefers the API.
- **Keys are never written to the repo.** The wizard writes them only to your shell profile, or to Claude Desktop's own secure settings.

---

## Development

```bash
cd mcp-server
npm install
npm run build
npx @modelcontextprotocol/inspector node dist/index.js
```

After changing the source, regenerate the server embedded in the plugin and commit the result — the Claude Code plugin runs this file, not the npm package:

```bash
cd mcp-server && npm run prepare-plugin
```

To try the plugin locally before publishing:

```
/plugin marketplace add /full/path/to/google-labs-claude
/plugin install google-labs
```

To cut a release, push a tag like `v0.1.0` — GitHub Actions publishes to npm and attaches the `.mcpb` bundle to the release.

---

## License

MIT — see [LICENSE](LICENSE).

google-labs-claude

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README