Agent orchestration & presence
Show live agent activity and coordinate edits without a task lock.
Start a session with get_guidelines and get_editor_state to identify the focused design and active page. report_agent_activity shows an external agent's current work location and action in Fanta. Give each agent a stable agent_id, a readable agent_name, and an action; optionally include page, exact node, world coordinates as [x,y], project_root, source_path, and workspace (canvas, variables, or code). Report each visible editing milestone and when moving between files or workspaces. A source path must stay inside the active Fanta project. Set active: false when finished. Unrefreshed reports expire after about 30 seconds.
Pass the same identity as batch_design.activity when editing canvas nodes. The operation batch previews one step at a time and updates the agent's focus; Fanta can follow that agent across canvas pages, component views, variables, and code. The Canvas workspace shows a pulsing edge while an agent reports activity for the open project. activity is optional for older clients; unidentified external edits use a shared fallback identity. Follow may change the visible page while a batch runs; an omitted parent still uses the page captured when the batch started.
{"label":"Build card","activity":{"agent_id":"designer-a","agent_name":"Merlyn","action":"Building card","workspace":"canvas"},"ops":[{"op":"set_props","id":"<node-id>","name":"Card"}]}If several agents work on one project, assign distinct nodes or source files and use one coordinator to review their screenshots and Git diffs. batch_design commits or rolls back as one undoable transaction, but activity reports do not lock nodes or files. The external MCP surface has no spawn_agent, agent_playbook, or task-lock tool.
The built-in Agent Panel uses design tools and reports its own identity and activity. Its profile and permission controls differ from the external MCP interface.