feat: support stderr logging in Jupyter notebook environments

Fixes #156

In Jupyter notebook environments, passing sys.stderr directly to subprocess
doesn't work as expected - the stderr output is lost and users have no
visibility into server errors or crashes.

This change:
- Adds _is_jupyter_environment() to detect when running in Jupyter/IPython
- In Jupyter mode, captures stderr as a pipe and reads it asynchronously
- Prints stderr output with red ANSI color for visibility in notebooks
- In non-Jupyter environments, behavior remains unchanged (direct stderr)

The fix ensures that MCP server error messages and crash logs are visible
in Jupyter notebooks, making debugging much easier.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: BabyChrist666

src/mcp/client/stdio.py

@@ -1,9 +1,10 @@
 import logging
 import os
+import subprocess
 import sys
 from contextlib import asynccontextmanager
 from pathlib import Path
-from typing import Literal, TextIO
+from typing import Callable, Literal, TextIO
 
 import anyio
 import anyio.lowlevel
@@ -24,6 +25,32 @@
 
 logger = logging.getLogger(__name__)
 
+
+def _is_jupyter_environment() -> bool:
+    """Detect if code is running in a Jupyter notebook environment.
+
+    In Jupyter environments, sys.stderr doesn't work as expected when passed
+    to subprocess, so we need to handle stderr differently.
+
+    Returns:
+        bool: True if running in Jupyter/IPython notebook environment
+    """
+    try:
+        # Check for IPython kernel
+        from IPython import get_ipython
+
+        ipython = get_ipython()
+        if ipython is not None:
+            # Check if it's a notebook kernel (not just IPython terminal)
+            if "IPKernelApp" in ipython.config:
+                return True
+            # Also check for ZMQInteractiveShell which indicates notebook
+            if ipython.__class__.__name__ == "ZMQInteractiveShell":
+                return True
+    except (ImportError, AttributeError):
+        pass
+    return False
+
 # Environment variables to inherit by default
 DEFAULT_INHERITED_ENV_VARS = (
     [
@@ -105,6 +132,12 @@ class StdioServerParameters(BaseModel):
 async def stdio_client(server: StdioServerParameters, errlog: TextIO = sys.stderr):
     """Client transport for stdio: this will connect to a server by spawning a
     process and communicating with it over stdin/stdout.
+
+    Args:
+        server: Parameters for the server process
+        errlog: TextIO stream for stderr output. In Jupyter environments,
+                stderr is captured and printed to work around Jupyter's
+                limitations with subprocess stderr handling.
     """
     read_stream: MemoryObjectReceiveStream[SessionMessage | Exception]
     read_stream_writer: MemoryObjectSendStream[SessionMessage | Exception]
@@ -115,16 +148,20 @@ async def stdio_client(server: StdioServerParameters, errlog: TextIO = sys.stder
     read_stream_writer, read_stream = anyio.create_memory_object_stream(0)
     write_stream, write_stream_reader = anyio.create_memory_object_stream(0)
 
+    # Detect Jupyter environment for stderr handling
+    is_jupyter = _is_jupyter_environment()
+
     try:
         command = _get_executable_command(server.command)
 
-        # Open process with stderr piped for capture
+        # Open process with stderr handling based on environment
         process = await _create_platform_compatible_process(
             command=command,
             args=server.args,
             env=({**get_default_environment(), **server.env} if server.env is not None else get_default_environment()),
             errlog=errlog,
             cwd=server.cwd,
+            capture_stderr=is_jupyter,
         )
     except OSError:
         # Clean up streams if process creation fails
@@ -177,9 +214,41 @@ async def stdin_writer():
         except anyio.ClosedResourceError:  # pragma: no cover
             await anyio.lowlevel.checkpoint()
 
+    async def stderr_reader():
+        """Read stderr from the process and output it appropriately.
+
+        In Jupyter environments, stderr is captured as a pipe and printed
+        to make it visible in the notebook output. In normal environments,
+        stderr is passed directly to sys.stderr.
+
+        See: https://github.com/modelcontextprotocol/python-sdk/issues/156
+        """
+        if not process.stderr:
+            return
+
+        try:
+            async for chunk in TextReceiveStream(
+                process.stderr,
+                encoding=server.encoding,
+                errors=server.encoding_error_handler,
+            ):
+                # In Jupyter, print to stdout with red color for visibility
+                # In normal environments, write to the provided errlog
+                if is_jupyter:
+                    # Use ANSI red color for stderr in Jupyter
+                    print(f"\033[91m{chunk}\033[0m", end="", flush=True)
+                else:
+                    errlog.write(chunk)
+                    errlog.flush()
+        except anyio.ClosedResourceError:
+            await anyio.lowlevel.checkpoint()
+
     async with anyio.create_task_group() as tg, process:
         tg.start_soon(stdout_reader)
         tg.start_soon(stdin_writer)
+        # Only start stderr reader if we're capturing stderr (Jupyter mode)
+        if is_jupyter and process.stderr:
+            tg.start_soon(stderr_reader)
         try:
             yield read_stream, write_stream
         finally:
@@ -232,19 +301,36 @@ async def _create_platform_compatible_process(
     env: dict[str, str] | None = None,
     errlog: TextIO = sys.stderr,
     cwd: Path | str | None = None,
+    capture_stderr: bool = False,
 ):
     """Creates a subprocess in a platform-compatible way.
 
     Unix: Creates process in a new session/process group for killpg support
     Windows: Creates process in a Job Object for reliable child termination
+
+    Args:
+        command: The executable command to run
+        args: Command line arguments
+        env: Environment variables for the process
+        errlog: TextIO stream for stderr (used when capture_stderr=False)
+        cwd: Working directory for the process
+        capture_stderr: If True, stderr is captured as a pipe for async reading.
+                       This is needed for Jupyter environments where passing
+                       sys.stderr directly doesn't work properly.
+
+    Returns:
+        Process with stdin, stdout, and optionally stderr streams
     """
+    # Determine stderr handling: PIPE for capture, or redirect to errlog
+    stderr_target = subprocess.PIPE if capture_stderr else errlog
+
     if sys.platform == "win32":  # pragma: no cover
-        process = await create_windows_process(command, args, env, errlog, cwd)
+        process = await create_windows_process(command, args, env, stderr_target, cwd)
     else:  # pragma: lax no cover
         process = await anyio.open_process(
             [command, *args],
             env=env,
-            stderr=errlog,
+            stderr=stderr_target,
             cwd=cwd,
             start_new_session=True,
         )

src/mcp/os/win32/utilities.py

@@ -5,7 +5,7 @@
 import subprocess
 import sys
 from pathlib import Path
-from typing import BinaryIO, TextIO, cast
+from typing import BinaryIO, TextIO, Union, cast
 
 import anyio
 from anyio import to_thread
@@ -66,7 +66,7 @@ class FallbackProcess:
     """A fallback process wrapper for Windows to handle async I/O
     when using subprocess.Popen, which provides sync-only FileIO objects.
 
-    This wraps stdin and stdout into async-compatible
+    This wraps stdin, stdout, and optionally stderr into async-compatible
     streams (FileReadStream, FileWriteStream),
     so that MCP clients expecting async streams can work properly.
     """
@@ -75,10 +75,12 @@ def __init__(self, popen_obj: subprocess.Popen[bytes]):
         self.popen: subprocess.Popen[bytes] = popen_obj
         self.stdin_raw = popen_obj.stdin  # type: ignore[assignment]
         self.stdout_raw = popen_obj.stdout  # type: ignore[assignment]
-        self.stderr = popen_obj.stderr  # type: ignore[assignment]
+        self.stderr_raw = popen_obj.stderr  # type: ignore[assignment]
 
         self.stdin = FileWriteStream(cast(BinaryIO, self.stdin_raw)) if self.stdin_raw else None
         self.stdout = FileReadStream(cast(BinaryIO, self.stdout_raw)) if self.stdout_raw else None
+        # Wrap stderr as async stream if it was captured as PIPE
+        self.stderr = FileReadStream(cast(BinaryIO, self.stderr_raw)) if self.stderr_raw else None
 
     async def __aenter__(self):
         """Support async context manager entry."""
@@ -99,12 +101,14 @@ async def __aexit__(
             await self.stdin.aclose()
         if self.stdout:
             await self.stdout.aclose()
+        if self.stderr:
+            await self.stderr.aclose()
         if self.stdin_raw:
             self.stdin_raw.close()
         if self.stdout_raw:
             self.stdout_raw.close()
-        if self.stderr:
-            self.stderr.close()
+        if self.stderr_raw:
+            self.stderr_raw.close()
 
     async def wait(self):
         """Async wait for process completion."""
@@ -133,7 +137,7 @@ async def create_windows_process(
     command: str,
     args: list[str],
     env: dict[str, str] | None = None,
-    errlog: TextIO | None = sys.stderr,
+    errlog: Union[TextIO, int, None] = sys.stderr,
     cwd: Path | str | None = None,
 ) -> Process | FallbackProcess:
     """Creates a subprocess in a Windows-compatible way with Job Object support.
@@ -150,7 +154,8 @@ async def create_windows_process(
         command (str): The executable to run
         args (list[str]): List of command line arguments
         env (dict[str, str] | None): Environment variables
-        errlog (TextIO | None): Where to send stderr output (defaults to sys.stderr)
+        errlog: Where to send stderr output. Can be a TextIO stream (like sys.stderr),
+                subprocess.PIPE (-1) for capturing stderr, or None.
         cwd (Path | str | None): Working directory for the subprocess
 
     Returns:
@@ -191,7 +196,7 @@ async def _create_windows_fallback_process(
     command: str,
     args: list[str],
     env: dict[str, str] | None = None,
-    errlog: TextIO | None = sys.stderr,
+    errlog: Union[TextIO, int, None] = sys.stderr,
     cwd: Path | str | None = None,
 ) -> FallbackProcess:
     """Create a subprocess using subprocess.Popen as a fallback when anyio fails.

tests/client/test_stdio.py

@@ -13,6 +13,7 @@
 from mcp.client.stdio import (
     StdioServerParameters,
     _create_platform_compatible_process,
+    _is_jupyter_environment,
     _terminate_process_tree,
     stdio_client,
 )
@@ -620,3 +621,81 @@ def sigterm_handler(signum, frame):
         f"stdio_client cleanup took {elapsed:.1f} seconds for stdin-ignoring process. "
         f"Expected between 2-4 seconds (2s stdin timeout + termination time)."
     )
+
+
+class TestJupyterStderrSupport:
+    """Tests for Jupyter notebook stderr logging support.
+
+    See: https://github.com/modelcontextprotocol/python-sdk/issues/156
+    """
+
+    def test_jupyter_detection_not_in_jupyter(self):
+        """Test that _is_jupyter_environment returns False when not in Jupyter."""
+        # In a normal Python environment (like pytest), this should return False
+        result = _is_jupyter_environment()
+        assert result is False, "Should not detect Jupyter in normal Python environment"
+
+    def test_jupyter_detection_handles_missing_ipython(self):
+        """Test that _is_jupyter_environment handles missing IPython gracefully."""
+        # This test verifies the ImportError handling works
+        # by calling the function when IPython may or may not be installed
+        result = _is_jupyter_environment()
+        # Should return False (not crash) regardless of IPython availability
+        assert isinstance(result, bool)
+
+    @pytest.mark.anyio
+    async def test_stderr_captured_in_process(self):
+        """Test that stderr output from a subprocess can be captured."""
+        # Create a script that writes to stderr
+        script = textwrap.dedent(
+            '''
+            import sys
+            sys.stderr.write("test error message\\n")
+            sys.stderr.flush()
+            # Exit immediately
+            sys.exit(0)
+            '''
+        )
+
+        server_params = StdioServerParameters(
+            command=sys.executable,
+            args=["-c", script],
+        )
+
+        # The stdio_client should handle this without hanging
+        with anyio.move_on_after(3.0) as cancel_scope:
+            async with stdio_client(server_params) as (read_stream, write_stream):
+                await anyio.sleep(0.5)  # Give process time to write and exit
+
+        assert not cancel_scope.cancelled_caught, "stdio_client should not hang on stderr output"
+
+    @pytest.mark.anyio
+    async def test_stderr_with_continuous_output(self):
+        """Test that continuous stderr output doesn't block the client."""
+        # Create a script that writes to stderr continuously then exits
+        script = textwrap.dedent(
+            '''
+            import sys
+            import time
+
+            for i in range(5):
+                sys.stderr.write(f"stderr line {i}\\n")
+                sys.stderr.flush()
+                time.sleep(0.1)
+
+            # Exit after writing
+            sys.exit(0)
+            '''
+        )
+
+        server_params = StdioServerParameters(
+            command=sys.executable,
+            args=["-c", script],
+        )
+
+        # The client should handle continuous stderr without blocking
+        with anyio.move_on_after(5.0) as cancel_scope:
+            async with stdio_client(server_params) as (read_stream, write_stream):
+                await anyio.sleep(1.0)  # Wait for stderr output
+
+        assert not cancel_scope.cancelled_caught, "stdio_client should handle continuous stderr output"