Filter response.mcp_approval_request.* substream events in _should_filter_mcp_chunk

The helper already handled mcp_approval_request items in the output_item.added
and output_item.done branches, but the substream event prefix check only matched
response.mcp_call.* and response.mcp_list_tools.*, allowing server-side approval
request progress events to leak to clients that cannot handle them.

Add the missing response.mcp_approval_request.* prefix check and add unit tests
for _should_filter_mcp_chunk covering all three substream event types.

Author: dprince

src/app/endpoints/responses.py

@@ -510,6 +510,7 @@ def _should_filter_mcp_chunk(
     if event_type and (
         event_type.startswith("response.mcp_call.")
         or event_type.startswith("response.mcp_list_tools.")
+        or event_type.startswith("response.mcp_approval_request.")
     ):
         output_index = getattr(chunk, "output_index", None)
         if output_index in server_mcp_output_indices:

tests/unit/app/endpoints/test_responses.py

@@ -14,6 +14,7 @@
 
 from app.endpoints.responses import (
     _is_server_mcp_output_item,
+    _should_filter_mcp_chunk,
     handle_non_streaming_response,
     handle_streaming_response,
     responses_endpoint_handler,
@@ -1425,3 +1426,141 @@ def test_file_search_call_returns_false(self) -> None:
         """Test file_search_call type returns False."""
         item: dict[str, Any] = {"type": "file_search_call"}
         assert _is_server_mcp_output_item(item, {"my-server"}) is False
+
+
+class TestShouldFilterMcpChunk:
+    """Tests for _should_filter_mcp_chunk helper."""
+
+    def test_filters_mcp_call_substream_events(self, mocker: MockerFixture) -> None:
+        """Test that response.mcp_call.* events are filtered for tracked indices."""
+        chunk = mocker.Mock()
+        chunk.output_index = 5
+        server_mcp_output_indices: set[int] = {5}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.mcp_call.in_progress",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is True
+        )
+
+    def test_filters_mcp_list_tools_substream_events(
+        self, mocker: MockerFixture
+    ) -> None:
+        """Test that response.mcp_list_tools.* events are filtered for tracked indices."""
+        chunk = mocker.Mock()
+        chunk.output_index = 3
+        server_mcp_output_indices: set[int] = {3}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.mcp_list_tools.in_progress",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is True
+        )
+
+    def test_filters_mcp_approval_request_substream_events(
+        self, mocker: MockerFixture
+    ) -> None:
+        """Test that response.mcp_approval_request.* events are filtered for tracked indices."""
+        chunk = mocker.Mock()
+        chunk.output_index = 7
+        server_mcp_output_indices: set[int] = {7}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.mcp_approval_request.in_progress",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is True
+        )
+
+    def test_does_not_filter_untracked_mcp_approval_request(
+        self, mocker: MockerFixture
+    ) -> None:
+        """Test that mcp_approval_request events for untracked indices pass through."""
+        chunk = mocker.Mock()
+        chunk.output_index = 7
+        server_mcp_output_indices: set[int] = {99}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.mcp_approval_request.in_progress",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is False
+        )
+
+    def test_does_not_filter_untracked_mcp_call(self, mocker: MockerFixture) -> None:
+        """Test that mcp_call events for untracked indices pass through."""
+        chunk = mocker.Mock()
+        chunk.output_index = 10
+        server_mcp_output_indices: set[int] = {5}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.mcp_call.completed",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is False
+        )
+
+    def test_filters_output_item_added_for_server_mcp(
+        self, mocker: MockerFixture
+    ) -> None:
+        """Test that output_item.added for server MCP items is filtered and tracked."""
+        item = mocker.Mock()
+        item.type = "mcp_approval_request"
+        item.server_label = "server-a"
+        chunk = mocker.Mock()
+        chunk.item = item
+        chunk.output_index = 2
+        server_mcp_output_indices: set[int] = set()
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.output_item.added",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is True
+        )
+        assert 2 in server_mcp_output_indices
+
+    def test_filters_output_item_done_for_server_mcp(
+        self, mocker: MockerFixture
+    ) -> None:
+        """Test that output_item.done for server MCP items is filtered and cleaned up."""
+        item = mocker.Mock()
+        item.type = "mcp_approval_request"
+        chunk = mocker.Mock()
+        chunk.item = item
+        chunk.output_index = 2
+        server_mcp_output_indices: set[int] = {2}
+        assert (
+            _should_filter_mcp_chunk(
+                chunk,
+                "response.output_item.done",
+                {"server-a"},
+                server_mcp_output_indices,
+            )
+            is True
+        )
+        assert 2 not in server_mcp_output_indices
+
+    def test_does_not_filter_non_mcp_event(self, mocker: MockerFixture) -> None:
+        """Test that non-MCP events pass through."""
+        chunk = mocker.Mock()
+        assert (
+            _should_filter_mcp_chunk(
+                chunk, "response.output_text.delta", {"server-a"}, set()
+            )
+            is False
+        )