fix(mcp): clean up failed HTTP connections

nanobot/agent/tools/mcp.py

@@ -975,11 +975,8 @@ async def connect_mcp_servers(
     from mcp.client.streamable_http import streamable_http_client
 
     async def open_single_server(
-        name: str, cfg: "MCPServerConfig"
+        name: str, cfg: "MCPServerConfig", server_stack: AsyncExitStack
-    ) -> tuple[str, AsyncExitStack | None]:
+    ) -> bool:
-        server_stack = AsyncExitStack()
-        await server_stack.__aenter__()
-
         try:
             transport_type = cfg.type
             if not transport_type:
@@ -991,8 +988,7 @@ async def connect_mcp_servers(
                     )
                 else:
                     logger.warning("MCP server '{}': no command or url configured, skipping", name)
-                    await server_stack.aclose()
+                    return False
-                    return name, None
 
             if transport_type in {"sse", "streamableHttp"}:
                 ok, error = validate_url_target(cfg.url)
@@ -1003,8 +999,7 @@ async def connect_mcp_servers(
                         _redact_url(cfg.url),
                         error,
                     )
-                    await server_stack.aclose()
+                    return False
-                    return name, None
 
             if transport_type == "stdio":
                 command, args, env = _normalize_windows_stdio_command(
@@ -1022,8 +1017,7 @@ async def connect_mcp_servers(
             elif transport_type == "sse":
                 if not await _probe_http_url(cfg.url):
                     logger.warning("MCP server '{}': {} unreachable, skipping", name, _redact_url(cfg.url))
-                    await server_stack.aclose()
+                    return False
-                    return name, None
 
                 def httpx_client_factory(
                     headers: dict[str, str] | None = None,
@@ -1050,8 +1044,7 @@ async def connect_mcp_servers(
             elif transport_type == "streamableHttp":
                 if not await _probe_http_url(cfg.url):
                     logger.warning("MCP server '{}': {} unreachable, skipping", name, _redact_url(cfg.url))
-                    await server_stack.aclose()
+                    return False
-                    return name, None
 
                 http_client = await server_stack.enter_async_context(
                     httpx.AsyncClient(
@@ -1067,8 +1060,7 @@ async def connect_mcp_servers(
                 )
             else:
                 logger.warning("MCP server '{}': unknown transport type '{}'", name, transport_type)
-                await server_stack.aclose()
+                return False
-                return name, None
 
             read = _filter_malformed_mcp_progress_notifications(read, name)
             session = await server_stack.enter_async_context(ClientSession(read, write))
@@ -1171,7 +1163,7 @@ async def connect_mcp_servers(
             logger.info(
                 "MCP server '{}': connected, {} capabilities registered", name, registered_count
             )
-            return name, server_stack
+            return True
 
         except Exception as e:
             hint = ""
@@ -1191,9 +1183,7 @@ async def connect_mcp_servers(
                     "only JSON-RPC to stdout and sends logs/debug output to stderr instead."
                 )
             logger.exception("MCP server '{}': failed to connect: {}", name, hint)
-            with suppress(Exception):
+            return False
-                await server_stack.aclose()
-            return name, None
 
     async def connect_single_server(
         name: str, cfg: "MCPServerConfig"
@@ -1203,30 +1193,30 @@ async def connect_mcp_servers(
         close_requested = asyncio.Event()
 
         async def own_connection() -> None:
-            stack: AsyncExitStack | None = None
             try:
-                _, stack = await open_single_server(name, cfg)
+                async with AsyncExitStack() as stack:
-                if not ready.done():
+                    connected = await open_single_server(name, cfg, stack)
-                    ready.set_result(stack is not None)
+                    if not ready.done():
-                if stack is not None:
+                        ready.set_result(connected)
-                    await close_requested.wait()
+                    if connected:
+                        await close_requested.wait()
             except BaseException as exc:
                 if not ready.done():
                     ready.set_exception(exc)
                 raise
-            finally:
-                if stack is not None:
-                    await stack.aclose()
 
         owner = asyncio.create_task(own_connection(), name=f"mcp:{name}")
         connection = _OwnedMCPConnection(owner, close_requested)
         try:
             connected = await ready
-        except BaseException:
+        except BaseException as exc:
             close_requested.set()
             owner.cancel()
             with suppress(BaseException):
                 await asyncio.shield(owner)
+            if isinstance(exc, asyncio.CancelledError) and not task_is_cancelling():
+                logger.warning("MCP server '{}': connection cancelled by server/SDK", name)
+                return name, None
             raise
         if not connected:
             await connection.aclose()

tests/tools/test_mcp_probe.py

@@ -5,10 +5,13 @@ import asyncio
 import socket
 from unittest.mock import MagicMock, patch
 
+import httpx
 import pytest
 
+from nanobot.agent.tools import mcp as mcp_mod
 from nanobot.agent.tools.mcp import _probe_http_url, connect_mcp_servers
 from nanobot.agent.tools.registry import ToolRegistry
+from nanobot.config.schema import MCPServerConfig
 from nanobot.security.network import configure_ssrf_whitelist
 
 _PROXY_ENV_VARS = ("HTTP_PROXY", "HTTPS_PROXY", "ALL_PROXY", "http_proxy", "https_proxy", "all_proxy")
@@ -171,6 +174,58 @@ async def test_connect_skips_unreachable_sse():
     assert len(registry._tools) == 0
 
 
+@pytest.mark.asyncio
+async def test_connect_isolates_streamable_http_status_failure(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    """A reachable endpoint returning HTTP 530 must not poison the event loop."""
+    async def _reachable(_url: str) -> bool:
+        return True
+
+    def _return_http_530(request: httpx.Request) -> httpx.Response:
+        return httpx.Response(530, text="cloudflare error 1033", request=request)
+
+    monkeypatch.setattr(mcp_mod, "validate_url_target", lambda _url: (True, ""))
+    monkeypatch.setattr(mcp_mod, "_probe_http_url", _reachable)
+    monkeypatch.setattr(
+        mcp_mod,
+        "PinnedDNSAsyncTransport",
+        lambda: httpx.MockTransport(_return_http_530),
+    )
+
+    loop = asyncio.get_running_loop()
+    previous_exception_handler = loop.get_exception_handler()
+    unhandled: list[BaseException] = []
+
+    def _capture_unhandled(_loop: asyncio.AbstractEventLoop, context: dict) -> None:
+        if isinstance(context.get("exception"), BaseException):
+            unhandled.append(context["exception"])
+
+    loop.set_exception_handler(_capture_unhandled)
+    try:
+        registry = ToolRegistry()
+        stacks = await asyncio.wait_for(
+            connect_mcp_servers(
+                {
+                    "cloudflare": MCPServerConfig(
+                        type="streamableHttp",
+                        url="https://mcp.example.com/mcp",
+                    )
+                },
+                registry,
+            ),
+            timeout=5.0,
+        )
+        await asyncio.sleep(0)
+
+        assert stacks == {}
+        assert registry.tool_names == []
+        assert unhandled == []
+        assert not any(task.get_name() == "mcp:cloudflare" for task in asyncio.all_tasks())
+    finally:
+        loop.set_exception_handler(previous_exception_handler)
+
+
 @pytest.mark.asyncio
 async def test_probe_not_called_for_stdio():
     """stdio transport should not be probed — it spawns a local process."""

tests/tools/test_mcp_tool.py

@@ -1082,10 +1082,22 @@ async def test_connect_mcp_servers_http_clients_reject_unsafe_redirect_targets(
 
 
 @pytest.mark.asyncio
+@pytest.mark.parametrize("failure_mode", ["exception", "cancellation"])
 async def test_connect_mcp_servers_one_failure_does_not_block_others(
     monkeypatch: pytest.MonkeyPatch,
+    failure_mode: str,
 ) -> None:
-    sessions = {"good": _make_fake_session(["demo"])}
+    bad_session = _make_fake_session([])
+
+    async def _cancel_initialize() -> None:
+        raise asyncio.CancelledError("cancelled by SDK")
+
+    if failure_mode == "cancellation":
+        bad_session.initialize = _cancel_initialize
+    sessions = {
+        "bad": bad_session,
+        "good": _make_fake_session(["demo"]),
+    }
 
     class _SelectiveClientSession:
         def __init__(self, read: object, _write: object) -> None:
@@ -1099,7 +1111,7 @@ async def test_connect_mcp_servers_one_failure_does_not_block_others(
 
     @asynccontextmanager
     async def _selective_stdio_client(params: object):
-        if params.command == "bad":
+        if params.command == "bad" and failure_mode == "exception":
             raise RuntimeError("boom")
         yield params.command, object()
 
@@ -1109,8 +1121,8 @@ async def test_connect_mcp_servers_one_failure_does_not_block_others(
     registry = ToolRegistry()
     stacks = await connect_mcp_servers(
         {
-            "good": MCPServerConfig(command="good"),
             "bad": MCPServerConfig(command="bad"),
+            "good": MCPServerConfig(command="good"),
         },
         registry,
     )
@@ -1121,6 +1133,36 @@ async def test_connect_mcp_servers_one_failure_does_not_block_others(
     assert set(stacks) == {"good"}
 
 
+@pytest.mark.asyncio
+async def test_connect_mcp_servers_propagates_external_cancellation(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    started = asyncio.Event()
+    closed = asyncio.Event()
+
+    @asynccontextmanager
+    async def _blocking_stdio_client(_params: object):
+        try:
+            started.set()
+            await asyncio.Event().wait()
+            yield object(), object()
+        finally:
+            closed.set()
+
+    monkeypatch.setattr(sys.modules["mcp.client.stdio"], "stdio_client", _blocking_stdio_client)
+
+    task = asyncio.create_task(
+        connect_mcp_servers({"slow": MCPServerConfig(command="slow")}, ToolRegistry())
+    )
+    await asyncio.wait_for(started.wait(), timeout=1.0)
+    task.cancel()
+
+    with pytest.raises(asyncio.CancelledError):
+        await task
+
+    await asyncio.wait_for(closed.wait(), timeout=1.0)
+
+
 @pytest.mark.asyncio
 async def test_connect_mcp_servers_streamable_http_uses_finite_timeout(
     fake_mcp_runtime: dict[str, object | None],