Connect via MCP
Use CutPro inside Claude, ChatGPT and other AI tools through the Model Context Protocol
MCP (Model Context Protocol) is an open standard that connects AI tools to external services. Connected to CutPro, the model generates images and videos with any top model (Veo, Kling, Seedance, Wan, Nano Banana, GPT Image and more), creates AI personas and puts them in videos, and clips long videos from start to finish, handing back the link. In clients with MCP Apps the results play right in the chat, and clients that support confirmations show the price and wait for your yes before any charge.
The local server uses an API key, which comes with the Gratuito, Básico, Plus, and Pro plans: create one in API keys and use it as CUTPRO_API_KEY. The remote server below works on every plan, with no key.
Local: Claude Code, Cursor, VS Code
The server runs on your machine through npx and exposes every v1 API endpoint as a tool, one per endpoint: image and video generation, AI personas, workspace, balance, videos and uploads, clipping, clips, channel and live monitoring, templates, renders, transcription, publishing and connections. Whatever the API can do, the model can do.
claude mcp add cutpro \
--env CUTPRO_API_KEY=$CUTPRO_API_KEY \
-- npx -y @cutpro/mcp@latestMulti-workspace key? Add CUTPRO_WORKSPACE_ID to the env as well.
An example
Once connected, just ask in plain language:
"Analyze this YouTube video, generate the clips, and send me the 3 best-rated ones already rendered."
The AI chains the tools by itself: analyze_video, create_submission, get_submission (until it finishes), list_clips, render_clip, get_render and download_render.
"Make a 5-second vertical video of a barista pouring latte art with Veo, then show me the price before you start."
The AI reads the models and prices with list_generation_models, tells you the total, and after your yes calls create_generation and get_generation until the video is ready. To animate an image it just made, it passes that image's output.image_id in image_ids.
Remote: ChatGPT and Claude.ai
The same server runs at https://mcp.cut.pro with full OAuth 2.1 (metadata discovery, dynamic client registration, PKCE and refresh tokens). It is the path for clients that run in the cloud and cannot execute npx.
Add the connector
In ChatGPT (Connectors) or Claude.ai (Settings, Connectors, Add custom connector), enter the URL https://mcp.cut.pro.
Sign in and approve
The client opens cut.pro. Sign in, pick the workspace the app may use and click Connect. Jobs use that workspace's credits, on any plan, and you can disconnect at any time in Connected apps.
Done
The client can now use CutPro tools in conversations.
Let the AI read these docs
For questions about the product, without running anything, the documentation is published as plain text in the format models read directly:
| Address | What it holds |
|---|---|
/docs/llms.txt | The index of every page, in all three languages |
/docs/llms-full.txt | The whole documentation in a single file |
Paste either one into a conversation and ask away. Any page also answers in Markdown if you swap the URL for .md, as in /docs/en/api-reference/quickstart.md.
Ask the AI to write the integration
If you want the code in your own project instead of running everything through chat, paste this into Cursor, Claude Code or whichever assistant you use:
Read https://cut.pro/docs/llms-full.txt and the spec at
https://api.cut.pro/api/v1/openapi.json.
Write a script that:
1. analyzes the video URL I give it (POST /clips/info) and tells me the cost
in credits before spending anything;
2. submits it for clipping and follows it until it finishes;
3. renders the highest-rated clip and gives me the MP4 link.
API rules: the key goes in the X-Api-Key header; every field is snake_case;
a resource's own id is always "id"; every error comes back as { code, extra },
so branch on the code and honour the Retry-After header on a 429.An API key reaches the workspace's credits and connected accounts. Keep it in an environment variable, and do not paste the key itself into the code you ask the AI to write.
Next step
Rather call the API directly?
The Quickstart has the whole flow in curl, Node.js and Python