diff --git a/docs/tool-guide.md b/docs/tool-guide.md
index a0b807d..ed54a58 100644
--- a/docs/tool-guide.md
+++ b/docs/tool-guide.md
@@ -36,6 +36,9 @@ Record numbering begins at 1 for the header; it counts parsed CSV records, not
 physical lines inside quoted values. Entirely blank lines are skipped by the parser.
 Empty input returns empty lists and zero rows. Malformed quoting returns `valid=False`
 and an error message. Extra cells are flagged but not counted as named columns.
+Header cells that are empty or whitespace-only are reported by zero-based position in
+`empty_header_indices`; a blank name is a warning, not an invalid document, and the
+original names are kept in `columns` unchanged.
 
 ## hash_text(text, algorithm="sha256")
 
diff --git a/src/mcp_tools_lab/tools/csv_tool.py b/src/mcp_tools_lab/tools/csv_tool.py
index 4fd3e19..9235197 100644
--- a/src/mcp_tools_lab/tools/csv_tool.py
+++ b/src/mcp_tools_lab/tools/csv_tool.py
@@ -8,7 +8,10 @@ def inspect_csv(text: str) -> dict:
     """Report rows, empty cells, and ragged records (record 1 is the header).
 
     Missing counts include absent trailing cells and whitespace-only cells.
-    Entirely blank records are skipped by Python's CSV reader.
+    Entirely blank records are skipped by Python's CSV reader. Header cells that
+    are empty or whitespace-only are reported by zero-based position in
+    ``empty_header_indices``; the original names are still kept in ``columns``,
+    and a blank name is a warning, not an invalid document.
     """
     try:
         rows = list(csv.reader(io.StringIO(text, newline=""), strict=True))
@@ -16,8 +19,10 @@ def inspect_csv(text: str) -> dict:
         return {"valid": False, "error": str(error)}
     if not rows:
         return {"valid": True, "columns": [], "row_count": 0,
-                "missing_values": [], "inconsistent_rows": []}
+                "missing_values": [], "inconsistent_rows": [],
+                "empty_header_indices": []}
     header, data = rows[0], rows[1:]
+    empty_header_indices = [index for index, name in enumerate(header) if not name.strip()]
     missing = [0] * len(header)
     inconsistent = []
     for record, row in enumerate(data, start=2):
@@ -30,4 +35,5 @@ def inspect_csv(text: str) -> dict:
             "missing_values": [
                 {"column": name, "index": index, "count": missing[index]}
                 for index, name in enumerate(header)
-            ], "inconsistent_rows": inconsistent}
+            ], "inconsistent_rows": inconsistent,
+            "empty_header_indices": empty_header_indices}
diff --git a/tests/test_csv_tool_empty_header.py b/tests/test_csv_tool_empty_header.py
new file mode 100644
index 0000000..a755334
--- /dev/null
+++ b/tests/test_csv_tool_empty_header.py
@@ -0,0 +1,36 @@
+from mcp_tools_lab.tools.csv_tool import inspect_csv
+
+
+def test_empty_header_indices_reports_positions():
+    result = inspect_csv("name,,age\nAda,20,x\nBob,30,y")
+    assert result["empty_header_indices"] == [1]
+
+
+def test_whitespace_only_header_reports_position():
+    result = inspect_csv("name, ,age\nAda,20,x")
+    assert result["empty_header_indices"] == [1]
+
+
+def test_empty_header_indices_lists_every_blank_column():
+    result = inspect_csv("a,,c,,e\n1,2,3,4,5")
+    assert result["empty_header_indices"] == [1, 3]
+
+
+def test_header_names_are_still_columns():
+    result = inspect_csv("name,,age\nAda,20,x")
+    assert result["columns"] == ["name", "", "age"]
+
+
+def test_empty_input_has_no_empty_header_indices():
+    assert inspect_csv("")["empty_header_indices"] == []
+
+
+def test_blank_header_is_still_valid():
+    assert inspect_csv("name,,age\nAda,20,x")["valid"] is True
+
+
+def test_malformed_csv_does_not_report_empty_header_indices():
+    # A document that will not parse has no header positions to name.
+    result = inspect_csv('name\n"unfinished')
+    assert result["valid"] is False
+    assert "empty_header_indices" not in result