proxy to timeout

src/wf_mcp/proxy/mounts.py

@@ -1,7 +1,9 @@
 from __future__ import annotations
 
+import asyncio
 import json
-from collections.abc import Callable
+import logging
+from collections.abc import Callable, Coroutine
 from dataclasses import dataclass
 from pathlib import Path
 from typing import Any, Generic, TypeVar
@@ -16,6 +18,8 @@ from ..shared.names import ProxyNamespace
 
 ProxyT = TypeVar("ProxyT")
 ProxyMountFactory = Callable[[ConnectionConfig, Path], "ProxyMount[ProxyT]"]
+_PROXY_LIST_TIMEOUT_SECONDS = 8.0
+logger = logging.getLogger(__name__)
 
 
 @dataclass(frozen=True, slots=True)
@@ -102,9 +106,10 @@ def create_proxy_mount(
     )
     transport = MCPConfigTransport(server_config, name_as_prefix=False)
     client = StatefulProxyClient(transport=transport, name=f"wf-mcp:{connection.id}")
-    proxy: FastMCP[Any] = FastMCPProxy(
+    proxy: FastMCP[Any] = ResilientFastMCPProxy(
         client_factory=client.new_stateful,
         name=f"Proxy-{connection.id}",
+        connection_id=connection.id,
     )
     proxy.add_transform(ProxyNamespace(connection.id))
     proxy.add_transform(ResourceLinkNamespace(connection.id))
@@ -113,3 +118,73 @@ def create_proxy_mount(
         fingerprint=connection_fingerprint(connection),
         proxy=proxy,
     )
+
+
+class ResilientFastMCPProxy(FastMCPProxy):
+    """FastMCP proxy that keeps discovery/listing best-effort per source.
+
+    FastMCP's aggregate provider skips providers that raise, but it still waits
+    for each mounted provider to finish listing. A dead stdio server can
+    therefore make top-level `tools/list` look broken for the whole broker. This
+    wrapper bounds list operations only; actual calls still use FastMCPProxy's
+    normal behavior and surface source failures. Timeout cancellation is also
+    the cleanup signal for FastMCP/MCP's stdio transport owner; this layer does
+    not launch or reap subprocesses directly.
+    """
+
+    def __init__(self, *, connection_id: str, **kwargs: Any) -> None:
+        super().__init__(**kwargs)
+        self._wf_mcp_connection_id = connection_id
+
+    async def list_tools(self, *, run_middleware: bool = True) -> Any:
+        return await _bounded_proxy_list(
+            super().list_tools(run_middleware=run_middleware),
+            connection_id=self._wf_mcp_connection_id,
+            operation="tools/list",
+        )
+
+    async def list_resources(self, *, run_middleware: bool = True) -> Any:
+        return await _bounded_proxy_list(
+            super().list_resources(run_middleware=run_middleware),
+            connection_id=self._wf_mcp_connection_id,
+            operation="resources/list",
+        )
+
+    async def list_resource_templates(self, *, run_middleware: bool = True) -> Any:
+        return await _bounded_proxy_list(
+            super().list_resource_templates(run_middleware=run_middleware),
+            connection_id=self._wf_mcp_connection_id,
+            operation="resources/templates/list",
+        )
+
+    async def list_prompts(self, *, run_middleware: bool = True) -> Any:
+        return await _bounded_proxy_list(
+            super().list_prompts(run_middleware=run_middleware),
+            connection_id=self._wf_mcp_connection_id,
+            operation="prompts/list",
+        )
+
+
+async def _bounded_proxy_list(
+    listing: Coroutine[Any, Any, Any],
+    *,
+    connection_id: str,
+    operation: str,
+    timeout_seconds: float = _PROXY_LIST_TIMEOUT_SECONDS,
+) -> Any:
+    """Return proxy list results or empty list for source/transport failures.
+
+    Only timeout and transport-ish failures are swallowed. Programming errors
+    should still escape to FastMCP's aggregate provider, which logs and skips the
+    mounted provider without hiding the bug from local tests.
+    """
+    try:
+        return await asyncio.wait_for(listing, timeout=timeout_seconds)
+    except (TimeoutError, OSError, ConnectionError) as exc:
+        logger.warning(
+            "Skipping %s for connection %s after listing failure: %s",
+            operation,
+            connection_id,
+            exc,
+        )
+        return []

tests/wf_mcp/test_proxy.py

@@ -12,6 +12,7 @@ from wf_mcp.broker import load_broker_config
 from wf_mcp.events import EventBus, InMemoryEventSink
 from wf_mcp.models import BrokerConfig, ConnectionConfig
 from wf_mcp.proxy import ProxyRuntime, create_proxy_client
+from wf_mcp.proxy.mounts import _bounded_proxy_list
 from wf_mcp.proxy.reload_events import (
     ProxyReloadResult,
     reload_change_events,
@@ -108,6 +109,39 @@ def test_proxy_lists_and_calls_upstream_tools() -> None:
     asyncio.run(run_proxy())
 
 
+def test_proxy_listing_degrades_when_one_source_hangs() -> None:
+    async def stuck_listing() -> list[Any]:
+        await asyncio.sleep(1)
+        return [{"name": "unreachable"}]
+
+    async def run_timeout() -> None:
+        result = await _bounded_proxy_list(
+            stuck_listing(),
+            connection_id="serena.default",
+            operation="tools/list",
+            timeout_seconds=0.01,
+        )
+        assert result == []
+
+    asyncio.run(run_timeout())
+
+
+def test_proxy_listing_degrades_when_one_source_has_transport_error() -> None:
+    async def broken_listing() -> list[Any]:
+        raise OSError("stdio process exited")
+
+    async def run_failure() -> None:
+        result = await _bounded_proxy_list(
+            broken_listing(),
+            connection_id="serena.default",
+            operation="tools/list",
+            timeout_seconds=1,
+        )
+        assert result == []
+
+    asyncio.run(run_failure())
+
+
 def test_proxy_registers_admin_tools_on_local_provider() -> None:
     config = BrokerConfig(
         store_root=local_temp_root() / "proxy_local_admin_store",

wf_mcp.config.json

@@ -8,8 +8,8 @@
       "enabled": true,
       "metadata": {
         "transport": "stdio",
-        "command": "pnpx",
-        "args": ["@upstash/context7-mcp"],
+        "command": "npx",
+        "args": ["-y", "@upstash/context7-mcp"],
         "env": {}
       }
     },
@@ -20,8 +20,8 @@
       "enabled": true,
       "metadata": {
         "transport": "stdio",
-        "command": "pnpx",
-        "args": ["@playwright/mcp@latest"],
+        "command": "npx",
+        "args": ["-y", "@playwright/mcp@latest"],
         "env": {}
       }
     },
@@ -32,8 +32,8 @@
       "enabled": true,
       "metadata": {
         "transport": "stdio",
-        "command": "pnpx",
-        "args": ["@modelcontextprotocol/server-everything"],
+        "command": "npx",
+        "args": ["-y", "@modelcontextprotocol/server-everything"],
         "env": {}
       }
     },
@@ -41,7 +41,7 @@
       "id": "serena.default",
       "server": "serena",
       "account": "default",
-      "enabled": false,
+      "enabled": true,
       "metadata": {
         "transport": "stdio",
         "command": "serena",