Creative Claw is an MCP plugin designed to embed a complete AI media studio into clients such as ChatGPT, Codex, Claude, Cursor, Grok Bot, Hermes Agent, and OpenClaw. Users can generate images, videos, and expressive speech directly in the conversation without switching platforms or configuring API keys, and reuse characters and brand assets to produce product marketing campaigns, UGC ads, and multi-shot films.
Main capabilities
- Image generation and editing: Gemini 3.1 Flash (Nano Banana 2) is recommended by default, balancing quality, speed, and cost; models such as GPT Image 2.5 and Seedream 5 Pro can also be used.
- Video generation: Supports models such as Seedance 2.5, Gemini Omni, and MiniMax H3 Max.
- Speech generation: Uses speech/elevenlabs-v4 by default for narration, dialogue, emotional expression, and multilingual speech; speech/xai-tts can also be used (28 built-in voices, inline/wrapped performance tags, G.711 telephony output). Supports designing synthetic voices from descriptions, or cloning a consented speaker's voice onto a character (Cartesia by default, ElevenLabs on request). The voice in existing recordings can also be changed via speech/cartesia-voice-changer.
- Characters and brand assets: Characters and brand assets can be saved and reused.
- Skills (Creative Workflows): The root skill creativeclaw handles routing of mixed or ambiguous requests, account issues, and the curated examples catalog; there are also 16 outcome-focused skills covering image, video, speech, music, and sound effect generation, characters and avatars, planning, films, product shoots, and UGC ads, editing existing media (cuts, reframing, subtitles, audio, intros and outros), long-video Reels, explicit HTML rendering, and feedback. Model guides are packaged as references within the outcome skills that use them; there are no model-specific skills.
- Model discovery and parameter queries: The current model catalog and parameters are obtained at runtime via list_models and get_model_params, rather than assuming a fixed catalog.
How to use
- ChatGPT: Install from the ChatGPT plugincatalog, log in, and state your needs in natural language. The plugin will declare the ChatGPT MCP dependency https://app.creativeclaw.co/mcp/chatgpt。
- Grok Bot and Cursor: Install via the Cursor Marketplace plugin system. In Grok Bot, open Settings → Plugins → Marketplace, or in Cursor open Customize → Plugins, search for Creative Claw and install it; browser authentication is completed on the first tool call. For local Cursor testing, symlink the plugin directory and reload Cursor.
- Hermes: Supports using the Creative Claw hosted MCP directly, running the corresponding command, or using the one-click Add to Hermes link. Until Hermes supports OAuth login for plugin-provided MCP entries, the two hermes mcp commands are the reliable installation path.
- OpenClaw: Includes a native OpenClaw manifest, consolidated skills, and ClawHub package metadata. Once published to ClawHub, it can be installed with the corresponding command. The first login opens Creative Claw's OAuth flow, with no need to copy API keys into configuration files.
- Agent skills (npx skills): Cross-client skills in the standard repository layout can be installed directly. The OpenAI Store separately installs the shared ChatGPT/Codex plugin and skill package; Store users do not need this command.
- Claude Desktop: Add the MCP configuration in claude_desktop_config.json. The first connection handles authentication via Clerk OAuth, with no API key required.
- Glama / catalog introspection: The glama-mcp/ directory provides a lightweight stdio MCP wrapper for Glama.ai catalog compatibility, exposing 9 core tool descriptions (how_to_connect, generate_image, generate_video, generate_speech, list_models, get_model_params, check_job, list_characters, get_credits_balance) and guiding users to connect to the full production MCP server https://app.creativeclaw.co/mcp。该包装器不执行实际生成，仅作为目录入口。
Use cases
- Generate videos in ChatGPT (since Sora was discontinued, ChatGPT has no built-in video generation; the plugin restores it).
- Create product ads, UGC ads, multi-shot films, product shoots.
- Add voiceover, dialogue, multilingual speech to videos, or clone your own voice.
- Edit existing media: cuts, reframing, subtitles, audio, intros and outros.
- Create long-form video Reels.
- Connect the plugin in ChatGPT dots and set up recurring tasks, such as weekly posts and Reels.
Limitations and notes
- Pay-as-you-go, no subscription: $10 buys 1,000 credits, new accounts get 100 free credits. You can request an estimate before generating. Current rates are at creativeclaw.co/pricing, and prices for each model within ChatGPT are at creativeclaw.co/chatgpt/pricing.
- Example costs: image 16 credits; 5-second video clip starting at 40 credits ($0.40); a 15-second product ad example with voiceover and subtitles is 461 credits ($4.61).
- Voice cloning requires recording or uploading a sample of your own voice (or a voice used with permission), saved as a reusable character, and can be used for video voiceover in more than 40 languages.
- Difference from Runway or HeyGen in ChatGPT: Creative Claw is a pay-as-you-go account that aggregates models from multiple providers (Seedance, Gemini Omni, MiniMax, Nano Banana, ElevenLabs, Cartesia), rather than a single vendor's model, and includes brand themes, reusable characters, voice cloning, and multi-shot assembly, without needing to buy a monthly plan first.
- Supported clients: ChatGPT, Codex, Claude (Code, Desktop, and web), Cursor, Grok Bot, Hermes Agent, and OpenClaw, all through the same account and MCP server https://app.creativeclaw.co/mcp。
- The repository is a marketplace structure containing one or more installable plugins. Two distributions are maintained: edit the canonical skill under plugins/creative-claw/skills/creativeclaw, put ChatGPT-specific routing differences in skill-variants/chatgpt/, then build the OpenAI package. The build script generates the plugin ZIP and skill downloads; the build process does not publish or commit plugin drafts.
