Before executing specific methods, read Craft scene commands and parameters, and execute according to the object selection, session, input replacement, and result checking described therein.
Connecting and executing
First check the current MCP tool list and editor_status. If create_node、run_command、inspect_document is present, use the actual tool JSON schema; do not guess parameters. Use list_commands to find the commands and parameters needed for this scene, then execute.
Enhanced Web MCP operations must happen on the same page that is open and ready. The official original Web 0.6.0 only has the interactive screen; the file service is not yet implemented; when these MCP tools are absent, do not pretend to have edited or saved. Standalone native figcraft-cli mcp is another session; switching must be explicitly announced. When the CLI is available, use figcraft-cli commands --json to check commands; do not install missing tools yourself.
Reference:FigCraft v0.6.0 MCP、CLI. Commands are subject to the currently loaded version.
Work steps
- First write out the user task, entry point, success path, cancel and error return paths; confirm whether the current page covers these states.
- Name the pages and key buttons, obtain node IDs via inspect_document; use list_commands to query proto.interaction.add and the flow and transition parameters of the current version.
- Connect the main actions one by one, specifying the target frame, trigger method, transition, and return behavior; do not fabricate backend responses that have not been provided.
- Actually run the prototype in the embedded page, click step by step from the entry point to the success and cancel states, and check the back stack and whether dead ends exist; the existence of static connections cannot replace click acceptance.
- Save and reopen to verify the connections; export the page overview and interaction notes, marking the actually clickable paths and the states that are only for explanation.