Creative Claw
On-brand creative studio for AI agents: images, video, audio, and 3D.
Links
README
From the repo.
Creative Claw
Generate on-brand media inside ChatGPT, Codex, Claude, Cursor, Grok Bot, Hermes, and OpenClaw.
Creative Claw is an MCP plugin that brings a full AI media studio into Grok Bot, Hermes Agent, OpenClaw, Cursor, Claude Code, Claude Desktop, ChatGPT, and Codex. Generate images, video, and expressive speech; reuse Characters and brand assets; and produce product campaigns, UGC ads, and multi-shot Films through one account. No API keys and no platform switching.
What You Get
MCP Server
One connection to Creative Claw's MCP server gives the skills live model discovery plus media generation, editing, brand, asset, Character, Film, and feedback tools:
| Category | Tools |
|---|---|
| Image | generate_image (generate, edit, or repeat across models for comparison), load_image |
| Video | generate_video, check_job |
| Speech | generate_speech |
| Media editing | remove_background, upscale_media, trim_video, scale_video, add_subtitles, extract_frames, merge_media |
| Models | list_models, get_model_params |
| Assets | search_assets, update_asset, delete_asset, upload_asset, import_media, get_upload_url, confirm_upload |
| Brand themes | get_theme, list_themes, update_theme, delete_theme |
| Characters | manage_character, list_characters, clone_voice, delete_character |
| Films | create_film_project, update_film_project, get_film_project, list_film_projects, assemble_film |
| Feedback | submit_feedback |
| Cost and account | estimate_generation; get_credits_balance and get_credits_link where the client exposes them |
Access 1,000+ production-ready AI models — FLUX, Gemini, Veo, Sora, Kling, Seedance, Hailuo, HeyGen, Recraft, ElevenLabs, and more — through a single unified account with usage-based pricing.
Skills (Creative Workflows)
The creativeclaw root skill routes mixed or unclear requests. Focused outcome skills own general image, video, and voice work plus Characters, planning, Films, product photoshoots, UGC ads, and feedback. Model specialists add exact prompting and reference guidance only after an outcome chooses Nano Banana 2, Nano Banana Pro, GPT Image 2.5 Flare or Sunburst, Seedream 5 Pro, Gemini Omni, Seedance 2.5, H3 Max, ElevenLabs v3, or xAI TTS.
Every OpenAI skill declares the ChatGPT MCP dependency at https://app.creativeclaw.co/mcp/chatgpt, matching the endpoint configured in the plugin draft. A focused skill can therefore activate directly without losing access to Creative Claw's tools.
Supported Models
Recommended image models
Default to Gemini 3.1 Flash (Nano Banana 2) for most image generation and editing. It is the most cost-efficient recommendation overall, balancing quality, speed, and cost; use a specialist model only when its specific strength matters.
| Model | Creative Claw ID | Best use |
|---|---|---|
| Nano Banana 2 | image/nano-banana-2 | Default and cost-efficient choice for most generation and editing |
| Nano Banana Pro | image/nano-banana-pro | Complex professional layouts, typography, and demanding composites |
| GPT Image 2.5 Flare | image/gpt-image-2.5-flare | Fast, high-quality everyday OpenAI image generation and editing |
| GPT Image 2.5 Sunburst | image/gpt-image-2.5-sunburst | Precision-focused generation, transparency, and tightly controlled edits |
| Seedream 5 Pro | image/seedream-5-pro | Premium product, fashion, and commercial imagery |
Recommended video models
| Model | Creative Claw ID | Best use |
|---|---|---|
| Gemini Omni | video/gemini-omni-flash | Default general video, native audio, references, and source-video work |
| Seedance 2.5 | video/seedance-2.5 | Premium, longer, reference-rich cinematic work |
| Seedance Mini | Discover with list_models | Economical drafts and iteration |
| MiniMax H3 Max | video/minimax-h3-max | Fast cinematic native-audio clips and multimodal references |
| H3 Max Fast | video/minimax-h3-max-turbo | Faster, lighter H3 Max iteration |
Recommended speech
Use speech/elevenlabs-v3 by default for narration, dialogue, emotional delivery, and multilingual speech. Use speech/xai-tts when its 28 built-in voices, exact inline/wrapping performance tags, or G.711 telephony output are the better fit. Creative Claw also supports consent-gated ElevenLabs Instant Voice Cloning as a separate Character workflow.
Use list_models and get_model_params at runtime rather than assuming a fixed catalog or reference limit.
Install
Grok Bot and Cursor
Creative Claw is packaged for the Cursor Marketplace used by Grok Bot's plugin system. After marketplace approval, open Settings → Plugins → Marketplace in Grok Bot or Customize → Plugins in Cursor, search for Creative Claw, install it, and complete browser authentication on the first tool call.
For local Cursor testing before publication, symlink the plugin directory and reload Cursor:
ln -s /path/to/creative-claw-marketplace/plugins/creative-claw ~/.cursor/plugins/local/creative-claw
Hermes Agent
Hermes supports the Creative Claw hosted MCP directly. Run:
hermes mcp add creative-claw --url https://app.creativeclaw.co/mcp --auth oauth
hermes mcp login creative-claw
Or use the one-click Add to Hermes link on a machine with Hermes installed. The portable plugin.json and mcp.json in the plugin root also make the package compatible with Agent Plugins v1 clients. Until Hermes supports OAuth login for plugin-supplied MCP entries, the two hermes mcp commands above are the reliable installation path.
OpenClaw
Creative Claw includes a native OpenClaw manifest, the consolidated skill, and ClawHub package metadata. Once the package is published to ClawHub, install it with:
openclaw plugins install clawhub:@creativeclaw/plugin
openclaw plugins enable creative-claw
openclaw mcp login creative-claw
openclaw gateway restart
The first login opens Creative Claw's OAuth flow; no API key needs to be copied into a config file.
Agent skills (npx skills)
The canonical cross-client skill remains in the standard repository layout, so skill-directory users can install it directly:
npx skills add CreativeClawCo/creative-claw-marketplace
The OpenAI Store installs the shared ChatGPT/Codex plugin and skill bundle separately; Store users do not need this command.
Claude Code
# 1. Add the marketplace
claude plugin marketplace add CreativeClawCo/creative-claw-marketplace
# 2. Install the plugin
claude plugin install creative-claw@creative-claw-marketplace
# 3. Authenticate — on first use, the MCP server will prompt you to sign in via Clerk OAuth
That's it. The plugin connects to Creative Claw's MCP server and installs the consolidated creativeclaw workflow skill.
Claude Desktop
Add to your MCP config (claude_desktop_config.json):
{
"mcpServers": {
"creative-claw": {
"type": "http",
"url": "https://app.creativeclaw.co/mcp"
}
}
}
No API keys needed — auth is handled via Clerk OAuth on first connection.
Glama / Directory Introspection
The glama-mcp/ directory provides a lightweight stdio MCP wrapper for Glama.ai directory compatibility. This wrapper exposes 9 core tool descriptions for discovery (how_to_connect, generate_image, generate_video, generate_speech, list_models, get_model_params, check_job, list_characters, get_credits_balance) that guide users to connect to the full production MCP server at https://app.creativeclaw.co/mcp.
The Glama wrapper does not perform actual generation—it serves as a directory entry point. To use Creative Claw's full generation capabilities, connect to the production HTTP MCP server as described above.
Run locally:
cd glama-mcp && npm install && npm start
Build Docker image:
docker build -t creative-claw-glama .
docker run -i creative-claw-glama
Quick Start
Talk to the host naturally:
"Generate a product photo of my headphones on a marble surface, golden hour lighting"
-> creativeclaw routes to its image workflow, picks the model, and generates
"Make a 15-second cinematic video of coffee being poured in slow motion"
-> creativeclaw routes to its video workflow, generates a reference image, then the video
"Edit this image — change the background to a beach sunset, keep the person unchanged"
-> creativeclaw routes to an edit model, preserves identity, swaps background
"Set up our brand — here's our website"
-> creativeclaw extracts colors/fonts/logos, uploads the assets, and saves a reusable theme
"I need a TikTok-style product video for this shoe" [attach image]
-> the UGC workflow plans the ad, creates reference frames, selects video and voice models, and assembles a first cut
"Create a reusable presenter from this portrait and clone my voice; I confirm it is mine"
-> the Character and consent-gated ElevenLabs voice-cloning workflows create and audition the reusable identity
"The video tool ignored my end frame—please report it"
-> the feedback workflow submits one actionable bug report with the attempted task and tool
Architecture
You -> Host + outcome skill (goal, approvals, creative direction)
|
Model specialist (prompt and reference details when selected)
|
MCP Tools (generation, assets, Characters, Films, feedback)
|
Creative Claw Server (model providers + R2 storage + Clerk auth)
|
Permanent media URLs (never expire)
No API keys. No CLI wrappers. No expiring URLs. Just skills + MCP.
Project Structure
This repo is a marketplace — it contains one or more installable plugins.
.claude-plugin/
marketplace.json # Marketplace manifest (required for Claude Code marketplace sync)
plugin.json # Root plugin metadata
.cursor-plugin/
marketplace.json # Cursor/Grok Bot marketplace manifest
plugins/
creative-claw/
.cursor-plugin/
plugin.json # Cursor/Grok Bot plugin manifest
.claude-plugin/
plugin.json # Plugin manifest (MCP server config)
.mcp.json # Grok Build/Codex MCP server config
plugin.json # Portable Agent Plugins v1 manifest (Hermes-compatible)
mcp.json # Portable Agent Plugins v1 MCP config
openclaw.plugin.json # Native OpenClaw manifest
skills/
creativeclaw/
SKILL.md # thin router for mixed and workspace requests
references/ # shared asset, theme, upload, and editing workflows
creativeclaw-*/ # outcome and recommended-model specialist skills
skill-variants/
chatgpt/
platform-upload.md # OpenAI Store routing for ChatGPT attachments and Codex local files
scripts/
build-skill-zips.sh # builds local archives into the ignored output folder
output/chatgpt-skills/ # local generated ZIPs, ignored by Git
evals/skill-routing-scenarios.md # activation and workflow regression suite
Maintaining the two distributions
Edit the canonical skill once under plugins/creative-claw/skills/creativeclaw. Put only ChatGPT-specific routing differences in skill-variants/chatgpt/platform-upload.md, then run:
./scripts/build-skill-zips.sh
The build validates every skill's metadata and MCP dependency, checks routing coverage and the regression suite, and creates deterministic archives under the Git-ignored output/chatgpt-skills/ folder. The marketplace source stays focused on direct installs, and generated ZIPs stay local. Nothing in the build publishes or submits a plugin draft.
Pricing
Usage-based — pay only for what you generate. No subscriptions, no commitments. Check creativeclaw.co for current rates.
Compatibility
- Grok Bot — via the Cursor marketplace plugin and hosted MCP server
- Cursor — via
.cursor-plugin/plugin.json - Claude Code — via
.claude-plugin/plugin.json - Claude Desktop — via MCP server config
- Codex and other skill-directory clients — via the canonical
creativeclawskill - OpenAI Store (ChatGPT + Codex) — build local ZIPs with
./scripts/build-skill-zips.sh, then useoutput/chatgpt-skills/creativeclaw-chatgpt-skill.zipand the focused skill ZIPs - Hermes Agent — via OAuth MCP setup and portable Agent Plugins v1 manifests
- OpenClaw — via
openclaw.plugin.jsonand ClawHub-ready package metadata
All use the same skills and connect to the same MCP server.
License
Apache-2.0
Built by Creative Claw Co.
Config for your environment
Use the endpoint URL below in your config. No API key — you connect directly.
Tool
OS
Config file: ~/.cursor/mcp.json
{
"mcpServers": {
"mcp-server": {
"url": "https://app.creativeclaw.co/mcp"
}
}
}Paste into mcpServers in the config file. Restart Cursor after saving.
If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.