[Bug]: MCP-FOR-UNITY transport lines (write FAIL / Client handler exited) pollute read_console error/warning results #1434

What happened?

MCP for Unity's own diagnostic lines show up in read_console results alongside the project's logs, including when filtering to errors/warnings. In one session, a types: ["error","warning"] read after a long-running operation returned only these:

MCP-FOR-UNITY: [IO] ✗ write FAIL  tag=response reqId=? ObjectDisposedException: Cannot access a disposed object.
MCP-FOR-UNITY: Suppressed duplicate command resent on a new connection; awaiting the in-flight result instead of running it twice.
MCP-FOR-UNITY: Client handler exited (remaining clients: 0)

This makes the common agent pattern "run something, then check the console for errors" unreliable. The agent has to recognise and discount the MCP's own transport noise every time, and real project errors can get crowded out by the count limit.

Reproduction steps

1. Run any tool call that outlives the frame timeout (e.g. a long execute_code), so the client reconnects.
2. Call read_console with types: ["error","warning"].
3. The MCP-FOR-UNITY transport lines appear among the results.

Expected / proposal: exclude MCP-FOR-UNITY-tagged entries from read_console by default, with an opt-in such as include_mcp_logs: true for debugging the bridge itself. Alternatively, log the expected transport-level reconnect/write failures at Log level rather than warning/error.

Environment

• Unity: 6000.3.8f1 (URP), Android build target
• MCP for Unity package: 10.2.1-beta.6 (git URL ?path=/MCPForUnity)
• Python server: mcpforunityserver, latest prerelease via uvx --prerelease explicit
• MCP client: Claude Code (CLI)
• Transport: stdio
• OS: Windows 11 Pro