Skip to content
Builder Studio
Get started

Model Context Protocol (MCP)

An open protocol that lets an AI agent discover and call another application's tools. Builder Studio runs an MCP server, so agents such as Claude Code, Codex and Cursor can work on your canvases.

An MCP server publishes a list of tools, each with a name, a description and the inputs it takes. A connected agent reads that list and calls the tools it needs, the way you would click through the product yourself.

How it works in Builder Studio

The Builder Studio MCP server is hosted at https://builder.studio/api/mcp. Nothing to install: add the address to your agent and sign in. An agent signs in one of two ways: with OAuth, where you approve access in your browser, or with a workspace API key created in workspace settings. Each tool needs its own permission, so an agent can call only what you granted.

The tools cover the work you do by hand:

  • Canvases: list, create, outline and screenshot canvases, write a page as HTML and CSS that lands as editable layers, and edit single layers.
  • Assets: import an image from a URL and place it on a page.
  • Sites: check a site before it publishes, manage CMS items, export the site as an Astro project and roll back to an earlier version.
  • Design systems: read, change, import and export the design graph.
  • Skills and Builder: read and save workspace skills, and post into a canvas's chat with Builder, the agent inside the product.

The base address serves a core set of tools. /api/mcp/full serves every tool your grant allows, for agents that need the whole surface.