Skip to content

Replace host operator with rescreen MCP-based computer use#43

Merged
DorianZheng merged 1 commit intomainfrom
feature/rescreen-mcp-host-computer-use
Mar 24, 2026
Merged

Replace host operator with rescreen MCP-based computer use#43
DorianZheng merged 1 commit intomainfrom
feature/rescreen-mcp-host-computer-use

Conversation

@DorianZheng
Copy link
Copy Markdown
Collaborator

Summary

  • Migrate from shell-script skills + custom Swift helper binary to rescreen MCP server for host desktop automation
  • Backend now speaks MCP protocol to rescreen directly, fixing broken postToPid keyboard events
  • Rename host-opshost-computer-use, gui-servicecomputer-use-service throughout
  • Add rescreen as git submodule, create host_computer_mcp.py MCP server for agent containers
  • Show TTL grants alongside permanent apps in host settings UI
  • Unfilter overview results so agents can see all windows for orientation

Test plan

  • make dev starts without errors
  • Create agent, configure allowed apps via Host settings
  • Agent can call rescreen_overview and see all windows
  • Agent can call rescreen_perceive with screenshot/accessibility modes
  • Agent can call rescreen_act with click/type/press actions (verify they actually work)
  • Approval flow: first request goes pending, admin approves, subsequent auto-approve
  • Host settings panel shows both permanent apps and TTL grants with time remaining

Migrate from shell-script skills + custom helper binary to rescreen
MCP server for host desktop automation. The backend now speaks MCP
protocol to rescreen directly, fixing broken postToPid keyboard
events and enabling element-based targeting.

- Add rescreen as git submodule at packages/backend/native/rescreen
- Create host_computer_mcp.py MCP server for agent containers
- Rename host-ops → host-computer-use, gui-service → computer-use-service
- Replace DarwinHostOperatorProvider with DarwinRescreenProvider
- Dynamic MCP config with WS URL + agent ID injection
- Show TTL grants alongside permanent apps in host settings UI
- Unfilter overview results so agents see all windows
- Remove dune-host-operator skill (replaced by MCP server)
@DorianZheng DorianZheng merged commit 03713d1 into main Mar 24, 2026
1 check failed
@DorianZheng DorianZheng deleted the feature/rescreen-mcp-host-computer-use branch March 24, 2026 08:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant