[BUG] preview_start spawns unlimited Node.js processes on Windows, consuming all available RAM
Bug Description
Triggering preview_start (via the "Start Preview" button in Claude Desktop or the mcp__Claude_Preview__preview_start MCP tool) spawns a brand new node.exe process every single time instead of reusing an existing one. On Windows, this results in 100+ Node.js processes accumulating in the background, consuming ALL available system RAM and causing repeated system crashes.
The tool description explicitly states "Reuses the server if already running" — this guarantee is completely broken on Windows.
Environment
- OS: Windows
- App: Claude Desktop
- Tool:
mcp__Claude_Preview__preview_start - Project type: Next.js 16 (
next dev, port 3000) .claude/launch.json: present and correctly configured
Steps to Reproduce
- Create
.claude/launch.jsonwith a Next.js dev server entry - Trigger
preview_start(via Claude Desktop UI or MCP tool call) - Trigger it again — even just once more
- Observe: a new
node.exeprocess is spawned instead of reusing the existing one - Repeat a few times → Task Manager shows 100+
node.exeprocesses
Expected Behavior
preview_start detects an already-running server on the target port and reuses it. The reused field in the response should return true on subsequent calls.
Actual Behavior
- Every call spawns a fresh
node.exeprocess (~31–33 MB each) - 100+
node.exeprocesses accumulate, consuming all remaining system RAM - System becomes completely unresponsive
- Machine crashed TWICE before the root cause was identified
reused: falseon every call, even repeated ones to the same named server
API Response Evidence
{
"serverId": "d9907155-d833-4c27-8632-c0e9368a7d1f",
"port": 3000,
"name": "Next.js Dev",
"reused": false
}
reused should be true on repeated calls — it never is.
Workaround
Kill all runaway processes via PowerShell:
Get-Process -Name "node" -ErrorAction SilentlyContinue | Stop-Process -Force
Then run npm run dev manually in a terminal instead of using preview_start.
Severity
Critical — causes repeated full system crashes on Windows by exhausting all available RAM. The bug is triggered by normal, expected usage of the feature.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗