Install

Connect an agent

Connect a trusted MCP client to give your agent access to an Open Canvas workspace.

Set up your MCP client

Use a client that can launch a local MCP server over standard input and output (stdio), such as the ones listed on the home page. Add the generated open-canvas entry to your client's MCP server configuration, then start or reconnect the client. On Windows the entry starts the command through cmd /c, because most MCP clients cannot launch a .cmd file directly.

Print the entry:

bash
open-canvas mcp-config

Example output

The launcher prints a server entry with this shape. Paths and ports shown here are examples; use the output generated for your installation.

json
{
  "mcpServers": {
    "open-canvas": {
      "command": "/absolute/path/to/.local/bin/open-canvas",
      "args": ["mcp"],
      "env": {
        "CAVAS_APP_PORT": "5190",
        "CAVAS_MCP_PORT": "4790"
      }
    }
  }
}

A typical agent workflow

An agent can discover an available canvas, read its document and layer tree, make an edit, then inspect the result:

  1. Call list_canvases and choose the canvas you intend to edit.
  2. Read it with get_tree or get_document.
  3. Make a change with create_artboard or create_nodes.
  4. Call get_screenshot when you need to inspect the artwork visually.

For a write, pass the revision from the last read as expectedRevision and its exact context with the project, file, and page IDs. Read the current state again after a change to confirm the result.

Pause access

Any process that can run the configured command can act with your local access to the editor. Only connect clients and agents you trust. Choose File → Pause agent access in Open Canvas to stop workspace and canvas access for agents. You can still edit and save in the browser. Resume it from that menu when you are ready.

Written for Open Canvas 0.8.0