Let an agent create media.Keep a project you can edit.
Connect Claude, ChatGPT or another MCP client to JsonCut. Your agent creates images and videos, brings in your media, looks at the frames it made and fills your templates, and everything it builds opens in Studio as normal layers.
Connect in a minute.Pick your client.
Claude users add JsonCut Studio from the Claude directory. ChatGPT, Claude Code, Codex, Cursor and Gemini CLI connect to the server URL with your JsonCut sign-in or an API key.
Connect ChatGPT
Sign in with your JsonCut accountAdd a custom MCP connection
In ChatGPT on the web, open Settings → Security and login → Developer mode. Open Plugins, choose + and add JsonCut with the server URL below. Menu names and availability can vary by account and workspace policy.
Connect with OAuth
Choose OAuth, follow the JsonCut sign-in flow and select Allow access. Do not use a JsonCut API key as a client ID or client secret. Return to ChatGPT and enable JsonCut in your conversation.
Check the connection
Start a new conversation with JsonCut enabled and ask: “List my JsonCut projects.” Check that the result matches your Studio workspace.
https://mcp.jsoncut.com/mcp“Create an editable launch image for my product and show me the result.”
The tools that matter.From a prompt to a reviewed file.
The server covers production work: projects, media, review, renders and templates. Billing and Studio internals are not exposed to agents.
Create and edit
Start an image or video, read its outline and change one part without replacing the rest.
jsoncut_create_projectjsoncut_get_project_outlinejsoncut_author_sceneBring real media
Import a file from a URL, upload bytes through a short-lived ticket, or hand the upload over to Studio.
jsoncut_import_media_urljsoncut_create_upload_ticketjsoncut_get_upload_handoffReview the result
Check the project, render frames of a scene to look at layout and timing, then render the file.
jsoncut_validate_projectjsoncut_review_scenejsoncut_render_projectReuse templates
Find a template, read its fields and render one version or a whole batch.
jsoncut_list_templatesjsoncut_get_template_inputsjsoncut_bulk_render_template
Long videos,one scene at a time.
A short clip is one call. For a longer film the agent keeps one project, reads its outline and writes and reviews one scene before the next. The 60-second brand film here is a project of that size; open it to see its scenes.
- Read the outline without loading the whole project
- Change one scene with its own timing
- Review frames before moving on
- Render the final version of the project
Let the agent build the template.Let a sheet make the next hundred.
Read before filling
The agent reads each field’s label, type, example and whether it is required, instead of guessing from a thumbnail.
One set of fields
The same fields work for a single render from chat and for a bulk run from CSV or JSON.
Still yours in Studio
Open the template in Studio whenever the design itself needs to change.
How templates work
Before you connect.Questions about JsonCut MCP.
Do I need an API key?
Not for Claude or ChatGPT: you sign in with your JsonCut account. Clients without OAuth, and unattended integrations, use an API key from Studio → Integrations → API keys.
Does every project need scenes?
No. Simple images and short videos stay flat. Scenes help when a longer video is built and reviewed one part at a time.
Can an agent upload my local files?
When the client can send file bytes, it uses a short-lived upload ticket. If the client cannot, the agent hands off to your Studio project, you upload the file there, and it continues with the same media.
Can the agent design freely?
Yes. It can create text, media, shapes, SVG, groups and keyframes directly. Library presets are optional starting points.
What about older V1 MCP clients?
The legacy server stays available at its own endpoint for compatibility. New clients should connect to the server shown in the setup guide.