fix(mcp): propagate HERMES_HOME override onto the MCP event loop (#44220)

* fix(mcp): propagate HERMES_HOME override onto the MCP event loop

Closes the known limit documented in #44007: tasks scheduled via
run_coroutine_threadsafe are created INSIDE the MCP loop thread, so they
copy that thread's context — a per-request profile scope (dashboard
?profile= endpoints, e.g. the MCP 'Test server' probe) silently vanished
for anything resolving get_hermes_home() inside the coroutine. Most
visible symptom: OAuth token-store paths (HERMES_HOME/mcp-tokens/)
resolved against the process home instead of the selected profile, so
testing an OAuth MCP cross-profile read the wrong tokens.

_run_on_mcp_loop now wraps scheduled coroutines with the caller's
context-local override (_wrap_with_home_override): set inside the task's
own context on the loop, reset on completion — task-local, so concurrent
calls carrying different scopes don't interfere, and the loop thread's
default context stays untouched. No-op (coroutine passes through
unwrapped) when no override is active, i.e. every non-dashboard caller.

web_server's probe comment updated from 'known limit' to 'covered'.

Tests: override propagation (direct + factory form), OAuth token-path
resolution on the loop, loop-context cleanliness after scoped calls,
no-op passthrough. 225 green across mcp_tool + unification suites.

* test(mcp): concurrent different-scope calls don't interfere
This commit is contained in:
Teknium
2026-06-11 04:37:01 -07:00
committed by GitHub
parent 3edd09a46f
commit 73dd584995
3 changed files with 188 additions and 4 deletions
+45 -1
View File
@@ -90,7 +90,7 @@ import sys
import threading
import time
from datetime import datetime
from typing import Any, Dict, List, Optional
from typing import Any, Coroutine, Dict, List, Optional
from urllib.parse import urlparse
logger = logging.getLogger(__name__)
@@ -2460,6 +2460,37 @@ def _ensure_mcp_loop():
_mcp_thread.start()
def _wrap_with_home_override(coro: "Coroutine") -> "Coroutine":
"""Carry the caller's context-local HERMES_HOME override into ``coro``.
Returns ``coro`` unchanged when no override is active. Otherwise wraps
it so the override is set inside the coroutine's own (task-local)
context on the MCP loop and reset when it completes — concurrent calls
carrying different scopes don't interfere.
"""
try:
from hermes_constants import (
get_hermes_home_override,
reset_hermes_home_override,
set_hermes_home_override,
)
home_override = get_hermes_home_override()
except Exception:
return coro
if not home_override:
return coro
async def _scoped():
token = set_hermes_home_override(home_override)
try:
return await coro
finally:
reset_hermes_home_override(token)
return _scoped()
def _run_on_mcp_loop(coro_or_factory, timeout: float = 30):
"""Schedule a coroutine on the MCP event loop and block until done.
@@ -2482,6 +2513,19 @@ def _run_on_mcp_loop(coro_or_factory, timeout: float = 30):
raise RuntimeError("MCP event loop is not running")
coro = coro_or_factory() if callable(coro_or_factory) else coro_or_factory
# Propagate the context-local HERMES_HOME override onto the MCP loop.
# Tasks scheduled via run_coroutine_threadsafe are created INSIDE the
# loop thread, so they copy the loop thread's context — not the
# scheduling thread's. A per-request profile scope (the dashboard's
# ?profile= endpoints, e.g. the MCP "Test server" probe) would silently
# vanish here: OAuth token stores and any other get_hermes_home()
# resolution inside the coroutine would read the process home instead
# of the selected profile's. Re-establish the override inside the
# task's own context (task-local — concurrent calls carrying different
# scopes don't interfere). No-op when no override is active.
coro = _wrap_with_home_override(coro)
future = safe_schedule_threadsafe(
coro, loop,
logger=logger,