fix(mcp): suppress benign 'Event loop is closed' noise on shutdown by acsezen · Pull Request #2538 · NousResearch/hermes-agent

acsezen · 2026-03-22T18:54:59Z

Problem

HTTP MCP servers (StreamableHTTP transport) print Unhandled exception in event loop: Event loop is closed on every session exit. This is pure stderr noise but alarming and may mask real errors.

Root cause: When _stop_mcp_loop() stops and closes the MCP background loop, httpx/httpcore async transports held by streamablehttp_client fire __del__ finalizers that call call_soon() on the now-dead loop. asyncio catches the RuntimeError and routes it to the loop's exception handler, which by default prints it to stderr. The connection teardown is a no-op at this point — safe to ignore.

Fix

Install a custom exception handler on the MCP event loop at creation time in _ensure_mcp_loop():

def _mcp_loop_exception_handler(loop, context):
    exc = context.get("exception")
    if isinstance(exc, RuntimeError) and "Event loop is closed" in str(exc):
        return  # benign shutdown race — suppress
    loop.default_exception_handler(context)

This silently drops RuntimeError('Event loop is closed') and forwards all other exceptions to the default handler, preserving visibility into real errors.

Testing

All 142 existing MCP tool tests pass (tests/tools/test_mcp_tool.py)
No new tests needed — this suppresses a stderr print path that has no observable side-effects

Pattern

Same class of async shutdown race fixed previously in:

model_tools.py (ab6abc2) — persistent event loop for tool handlers
agent/auxiliary_client.py — stale async client cache invalidation

When the MCP background loop is stopped and closed, httpx/httpcore async transports held by streamablehttp_client fire __del__ finalizers that call call_soon() on the now-dead loop. asyncio catches the resulting RuntimeError and routes it to the loop's exception handler, which by default prints 'Unhandled exception in event loop: Event loop is closed' on stderr. Fix: install a custom exception handler on the MCP event loop at creation time that silently drops RuntimeError('Event loop is closed') — a benign shutdown race — and forwards all other exceptions to the default handler. Root cause: Python's GC runs transport __del__ finalizers after the loop is closed. The transports try to schedule cleanup via call_soon(), which raises because the loop is already dead. The connection teardown is a no-op at this point so the error is safe to suppress. Matches the pattern used in model_tools.py (ab6abc2) and auxiliary_client.py (stale async client cache invalidation) for the same class of async shutdown races. Fixes NousResearch#2521

…t loop handler, OAuth non-blocking Four fixes for MCP server stability issues reported by community member (terminal lockup, zombie processes, escape sequence pollution, startup hang): 1. MCP reload timeout guard (cli.py): _check_config_mcp_changes now runs _reload_mcp in a separate daemon thread with a 30s hard timeout. Previously, a hung MCP server could block the process_loop thread indefinitely, freezing the entire TUI (user can type but nothing happens, only Ctrl+D/Ctrl+\ work). 2. MCP stdio subprocess PID tracking (mcp_tool.py): Tracks child PIDs spawned by stdio_client via before/after snapshots of /proc children. On shutdown, _stop_mcp_loop force-kills any tracked PIDs that survived the SDK's graceful SIGTERM→SIGKILL cleanup. Prevents zombie MCP server processes from accumulating across sessions. 3. MCP event loop exception handler (mcp_tool.py): Installs _mcp_loop_exception_handler on the MCP background event loop — same pattern as the existing _suppress_closed_loop_errors on prompt_toolkit's loop. Suppresses benign 'Event loop is closed' RuntimeError from httpx transport __del__ during MCP shutdown. Salvaged from PR #2538 (acsezen). 4. MCP OAuth non-blocking (mcp_oauth.py): Replaces blocking input() call in _wait_for_callback with OAuthNonInteractiveError raise. Adds _is_interactive() TTY detection. In non-interactive environments, build_oauth_auth() still returns a provider (cached tokens + refresh work), but the callback handler raises immediately instead of blocking the MCP event loop for 120s. Re-raises OAuth setup failures in _run_http so failed servers are reported cleanly without blocking others. Salvaged from PRs #4521 (voidborne-d) and #4465 (heathley). Closes #2537, closes #4462 Related: #4128, #3436

…t loop handler, OAuth non-blocking (#4757) Four fixes for MCP server stability issues reported by community member (terminal lockup, zombie processes, escape sequence pollution, startup hang): 1. MCP reload timeout guard (cli.py): _check_config_mcp_changes now runs _reload_mcp in a separate daemon thread with a 30s hard timeout. Previously, a hung MCP server could block the process_loop thread indefinitely, freezing the entire TUI (user can type but nothing happens, only Ctrl+D/Ctrl+\ work). 2. MCP stdio subprocess PID tracking (mcp_tool.py): Tracks child PIDs spawned by stdio_client via before/after snapshots of /proc children. On shutdown, _stop_mcp_loop force-kills any tracked PIDs that survived the SDK's graceful SIGTERM→SIGKILL cleanup. Prevents zombie MCP server processes from accumulating across sessions. 3. MCP event loop exception handler (mcp_tool.py): Installs _mcp_loop_exception_handler on the MCP background event loop — same pattern as the existing _suppress_closed_loop_errors on prompt_toolkit's loop. Suppresses benign 'Event loop is closed' RuntimeError from httpx transport __del__ during MCP shutdown. Salvaged from PR #2538 (acsezen). 4. MCP OAuth non-blocking (mcp_oauth.py): Replaces blocking input() call in _wait_for_callback with OAuthNonInteractiveError raise. Adds _is_interactive() TTY detection. In non-interactive environments, build_oauth_auth() still returns a provider (cached tokens + refresh work), but the callback handler raises immediately instead of blocking the MCP event loop for 120s. Re-raises OAuth setup failures in _run_http so failed servers are reported cleanly without blocking others. Salvaged from PRs #4521 (voidborne-d) and #4465 (heathley). Closes #2537, closes #4462 Related: #4128, #3436

teknium1 · 2026-04-03T09:30:24Z

Merged via PR #4757 (MCP stability fix pack). Your exception handler fix was incorporated directly — thank you @acsezen!

…t loop handler, OAuth non-blocking (NousResearch#4757) Four fixes for MCP server stability issues reported by community member (terminal lockup, zombie processes, escape sequence pollution, startup hang): 1. MCP reload timeout guard (cli.py): _check_config_mcp_changes now runs _reload_mcp in a separate daemon thread with a 30s hard timeout. Previously, a hung MCP server could block the process_loop thread indefinitely, freezing the entire TUI (user can type but nothing happens, only Ctrl+D/Ctrl+\ work). 2. MCP stdio subprocess PID tracking (mcp_tool.py): Tracks child PIDs spawned by stdio_client via before/after snapshots of /proc children. On shutdown, _stop_mcp_loop force-kills any tracked PIDs that survived the SDK's graceful SIGTERM→SIGKILL cleanup. Prevents zombie MCP server processes from accumulating across sessions. 3. MCP event loop exception handler (mcp_tool.py): Installs _mcp_loop_exception_handler on the MCP background event loop — same pattern as the existing _suppress_closed_loop_errors on prompt_toolkit's loop. Suppresses benign 'Event loop is closed' RuntimeError from httpx transport __del__ during MCP shutdown. Salvaged from PR NousResearch#2538 (acsezen). 4. MCP OAuth non-blocking (mcp_oauth.py): Replaces blocking input() call in _wait_for_callback with OAuthNonInteractiveError raise. Adds _is_interactive() TTY detection. In non-interactive environments, build_oauth_auth() still returns a provider (cached tokens + refresh work), but the callback handler raises immediately instead of blocking the MCP event loop for 120s. Re-raises OAuth setup failures in _run_http so failed servers are reported cleanly without blocking others. Salvaged from PRs NousResearch#4521 (voidborne-d) and NousResearch#4465 (heathley). Closes NousResearch#2537, closes NousResearch#4462 Related: NousResearch#4128, NousResearch#3436

teknium1 mentioned this pull request Apr 3, 2026

fix(mcp): stability fix pack — reload timeout, shutdown cleanup, event loop handler, OAuth non-blocking #4757

Merged

teknium1 closed this Apr 3, 2026

acsezen deleted the fix/mcp-event-loop-closed-shutdown branch April 3, 2026 09:55

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(mcp): suppress benign 'Event loop is closed' noise on shutdown#2538

fix(mcp): suppress benign 'Event loop is closed' noise on shutdown#2538
acsezen wants to merge 1 commit into
NousResearch:mainfrom
acsezen:fix/mcp-event-loop-closed-shutdown

acsezen commented Mar 22, 2026

Uh oh!

teknium1 commented Apr 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

acsezen commented Mar 22, 2026

Problem

Fix

Testing

Pattern

Uh oh!

teknium1 commented Apr 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants