/clear does not terminate previous stdio MCP server child processes (unlike /mcp reload )

[robgurr]

Description

When reloading MCP servers with /mcp reload, the CLI correctly terminates the existing stdio MCP server child process before starting a new one. However, running /clear (start fresh session) spawns a new stdio MCP server process without terminating the previous session's MCP server process, leaking orphaned processes over time.

Steps to reproduce

1. Start a session with a stdio-based MCP server configured (e.g. a custom *.exe MCP server).
2. Confirm the server process is running (e.g. via Task Manager / Get-Process).
3. Run /clear.
4. Observe: a new instance of the MCP server process starts, but the previous one is still running.
5. Repeat /clear multiple times — orphaned process count grows unbounded.
6. Contrast with /mcp reload, which correctly stops the old process before starting the new one.

Expected behavior

/clear should terminate/clean up MCP server child processes from the previous session (or reuse them) before/while starting the new session's servers, matching the behavior of /mcp reload.

Actual behavior

Old MCP stdio server processes are left running indefinitely after /clear, accumulating over repeated use (observed 22+ orphaned exe instances in one session).

Environment

• Copilot CLI version: 1.0.82
• OS: Windows
• MCP server type: stdio-based custom MCP server