Agents on the canvas

Three agents edit three frames in parallel, each with its own cursor and avatar.

Agents show up on the canvas the way teammates do.

• Cursor. When an agent reads or edits an element, its cursor glides to it, labelled with its name: Claude for the chat, Claude · subagent 2 for a sub-agent it started, MCP agent for an agent connected over MCP.
• Selection. The elements it is working on get an outline in its colour.
• Avatar. Each agent session gets its own shape, colour and initial, shown at the top right next to the people on the canvas, with a sparkle badge. Hover it to see what the agent is doing.
• Parallel work. Several agents at once means several cursors at once.
• Idle. A few seconds after an agent stops, its cursor, outline and avatar disappear.

Every collaborator on the canvas sees them, not only the person who started the agent.

With an MCP agent

Nothing to turn on. Every edit call that reads or changes a node moves the agent’s cursor there. Send the same x-feega-session header (a UUID) on every call to keep one cursor per agent session; a different session gets a different cursor and avatar.

edit { "op": "set_props", "target": "<frameId>", "input": { "clip_id": "title", "props": { "text": "Slow mornings, made by hand." } } }
edit { "op": "move_node", "input": { "nodeId": "<frameId>", "x": 760, "y": 480 } }