First off, this is a genuinely impressive project. The unified UIA + CDP + Vision architecture in a single MCP server is the right call. These bugs found during real use — reporting them to help make it even better.
1. type_text / clipboard write timeout
Input operations hang for text over ~20 bytes, triggering 120s MCP timeout. Short text (<10 chars) works fine. Both input type_text and clipboard write have the same behavior. Other tools (ping, exec, mouse_click) work during the same session.
2. ui-sight-windows.exe crashes
Intermittent exit code 1, ~30% of calls. Once it crashes, both vision window_layout and window screenshot fail until automata-agent restart.
3. browser ensure creates blank Edge window
Launches Edge with no user profile (no bookmarks/cookies/settings). CDP connection then times out after 30s. Should reuse the existing default profile.
4. MCP session lost after agent restart
Killing and restarting automata-agent breaks all existing Streamable HTTP sessions. Client gets stuck in not connected / unreachable state until manual reconnect.
5. workflow list returns empty
workflow list MCP tool returns []. Works fine via resources list (shows 35+ workflows).
6. find_elements returns empty
desktop find_elements with valid selectors (e.g. >> [role=button]) returns []. element_tree on the same window finds the elements.
Environment: Win11 zh-CN, UI Automata v0.1.1, WSL2 mirrored networking, Streamable HTTP on :3100.
Keep pushing — this project has the right foundation to become the go-to Windows automation toolkit.
First off, this is a genuinely impressive project. The unified UIA + CDP + Vision architecture in a single MCP server is the right call. These bugs found during real use — reporting them to help make it even better.
1. type_text / clipboard write timeout
Input operations hang for text over ~20 bytes, triggering 120s MCP timeout. Short text (<10 chars) works fine. Both
input type_textandclipboard writehave the same behavior. Other tools (ping, exec, mouse_click) work during the same session.2. ui-sight-windows.exe crashes
Intermittent exit code 1, ~30% of calls. Once it crashes, both
vision window_layoutandwindow screenshotfail until automata-agent restart.3. browser ensure creates blank Edge window
Launches Edge with no user profile (no bookmarks/cookies/settings). CDP connection then times out after 30s. Should reuse the existing default profile.
4. MCP session lost after agent restart
Killing and restarting automata-agent breaks all existing Streamable HTTP sessions. Client gets stuck in
not connected/unreachablestate until manual reconnect.5. workflow list returns empty
workflow listMCP tool returns[]. Works fine viaresources list(shows 35+ workflows).6. find_elements returns empty
desktop find_elementswith valid selectors (e.g.>> [role=button]) returns[].element_treeon the same window finds the elements.Environment: Win11 zh-CN, UI Automata v0.1.1, WSL2 mirrored networking, Streamable HTTP on :3100.
Keep pushing — this project has the right foundation to become the go-to Windows automation toolkit.