Several Figma files, several agents, one server
Open the Canvashand plugin in every file. Each agent or subagent works in its own file, in parallel, with its own change sessions and rollback journal.
Open the Canvashand plugin in every Figma file you work on, and each agent or subagent can take its own file. They run at the same time, and each file keeps its own change sessions and rollback journal.
The problem
Agent teams are natural for design work. One agent builds the design system. Another builds the onboarding flow. A third builds settings. They share a goal but not a canvas, so they should not block each other.
With one connected file at a time, they did. Agents queued behind whichever file was connected, or you swapped files by hand between runs. That wastes the main benefit of running agents in parallel.
The cost shows up fastest on real projects. A design system has to exist before screens can use its components, but onboarding and settings flows can start from an agreed spec at the same moment. When agents cannot work at once, the whole job takes as long as the sum of its parts, and you sit in the middle moving files around.
How it works
Open the plugin in as many files as you like. Every file's plugin joins the same Canvashand server on port 21994 (or the next free port up to 21998), so there are no extra ports to configure. With one file connected, tools work exactly as before.
With several files connected:
list_filesshows each file'sfileIdand name.select_filesets the default file for the current session, byfileIdor exact name.- Any tool accepts an optional
fileIdto target one file for that call. That is how parallel subagents in one session each work in their own file.
Each file keeps its own change sessions, its own rollback journal, and its own credit lease. One agent's rollback never touches another file's work. A file's ID is saved in the file, so reopening the plugin reconnects it under the same ID.
The same setup works across separate sessions. Two Claude Code windows, or Claude Code and Codex, can each work in their own file against the one server.
A session started in Onboarding does not capture a write made in Settings, and rollback_changes in one file deletes only what that file's session created. Every file's lease draws on the same account quota, so parallel agents share your plan's calls.
You still decide how to split the work. Give each agent a clear file and a clear brief, and name the file in the prompt so the agent passes the right fileId.
If you work alone, the same setup still helps. You can keep one agent on the design system while you review another agent's screens, without stopping either of them to switch files. Nothing about the server changes between one file and ten; you only open the plugin in more of them.
Try it
Open the plugin from Plugins > Development > Canvashand in each file. Then ask your agent:
Use three subagents in parallel, one per open Figma file from list_files: the design system in 'Design system', the onboarding flow in 'Onboarding', and the settings screens in 'Settings'.
To get the latest CLI and plugin, run:
canvashand upgrade
New to Canvashand? Start with Why we built Canvashand, or follow the setup page. For three starter prompts and how a design session runs, read Run AI agents and subagents in Figma.