Bitgrain ships a Model Context Protocol server, which lets an agent you control drive the studio: open a template, add and update layers, apply effects, render a page, export the result.
The distinction matters and it is worth being precise about. The effects engine contains no models. Nothing in the rendering pipeline guesses, generates or infers anything, and the output of a given document is still purely a function of its settings.
What the MCP server does is expose that deterministic engine as a set of tools. If you choose to point an agent at it, the agent is choosing which levers to pull, exactly as you would. It is not generating imagery, and the same document rendered by hand and by an agent produces identical pixels.
Browse and open templates, create documents, add and reorder layers, update text and imagery, apply and configure effects, render individual pages or a whole document, and fetch download URLs for the results.
Rendering requires a paid plan, the same as it does in the browser. Everything up to that point, including browsing templates and building a document, is free.
The server is added to an MCP-capable client the same way any other server is, and the first tool call prompts a sign in through the browser so the server acts on your account rather than on an anonymous one.
Sessions are per account. Documents an agent creates appear in your workspace and can be opened, edited and exported by hand afterwards, which is usually how it goes: the agent does the repetitive assembly and a person makes the judgement calls.
No. The engine contains no models and generates nothing. The server exposes deterministic tools that an agent you control can call, in exactly the way you would call them yourself.
Browsing templates, reading design skills and building documents are free. Rendering needs a paid plan, the same as it does in the browser.
In your own workspace, openable and editable by hand afterwards.