diff --git a/sentry_sdk/integrations/mcp.py b/sentry_sdk/integrations/mcp.py
index 5fc186f505..a3523266de 100644
--- a/sentry_sdk/integrations/mcp.py
+++ b/sentry_sdk/integrations/mcp.py
@@ -270,6 +270,19 @@ def _extract_text_from_content_blocks(content_blocks: "Any") -> "Any":
     return " ".join(texts) if texts else content_blocks
 
 
+def _tool_result_is_error(result: "Any") -> bool:
+    """
+    Return whether a tools/call result reports a tool error.
+
+    A dict uses the wire key ``isError`` (what mcp 2.x middleware observes).
+    A result object uses the ``isError`` attribute (``CallToolResult`` on mcp
+    1.x, and a non-dict result the v2 middleware can see).
+    """
+    if isinstance(result, dict):
+        return result.get("isError") is True
+    return getattr(result, "isError", None) is True
+
+
 async def _tool_handler_wrapper(
     func: "Callable[..., Awaitable[Union[CallToolResult, InputRequiredResult]]]",
     original_args: "tuple[Any, ...]",
@@ -364,6 +377,9 @@ async def _tool_handler_wrapper(
                     _capture_exception(e)
                 raise
 
+            if _tool_result_is_error(result):
+                _set_span_data_attribute(span, SPANDATA.ERROR_TYPE, "tool_error")
+
             if result is None:
                 return result
 
@@ -461,6 +477,9 @@ async def _instrument_v2_tool_call(
                     _capture_exception(e)
                 raise
 
+            if _tool_result_is_error(result):
+                _set_span_data_attribute(span, SPANDATA.ERROR_TYPE, "tool_error")
+
             if not isinstance(result, dict):
                 return result
 
diff --git a/tests/integrations/mcp/test_mcp.py b/tests/integrations/mcp/test_mcp.py
index 15432d8d84..cee93751cc 100644
--- a/tests/integrations/mcp/test_mcp.py
+++ b/tests/integrations/mcp/test_mcp.py
@@ -63,7 +63,7 @@ async def __call__(self, *args, **kwargs):
         TextResourceContents,
     )
 else:
-    from mcp.types import GetPromptResult, PromptMessage, TextContent
+    from mcp.types import CallToolResult, GetPromptResult, PromptMessage, TextContent
 
 from mcp.server.sse import SseServerTransport
 from mcp.server.streamable_http_manager import StreamableHTTPSessionManager
@@ -788,6 +788,51 @@ def failing_tool(tool_name, arguments):
         assert span["tags"]["status"] == "internal_error"
 
 
+@pytest.mark.asyncio
+async def test_tool_is_error_result_sets_error_type(sentry_init, capture_events, stdio):
+    sentry_init(
+        integrations=[MCPIntegration()],
+        default_integrations=False,
+        traces_sample_rate=1.0,
+        trace_lifecycle="static",
+    )
+
+    if IS_MCP_V2:
+        from mcp.server import MCPServer
+
+        mcp_server = MCPServer("test-server")
+
+        @mcp_server.tool()
+        def bad_tool() -> str:
+            raise ValueError("bad input")
+
+        server = mcp_server._lowlevel_server
+    else:
+        server = Server("test-server")
+
+        @server.call_tool()
+        def bad_tool(tool_name, arguments):
+            return CallToolResult(
+                content=[TextContent(type="text", text="bad input")],
+                isError=True,
+            )
+
+    events = capture_events()
+    with start_transaction(name="mcp tx"):
+        result = await stdio(
+            server,
+            method="tools/call",
+            params={"name": "bad_tool", "arguments": {}},
+            request_id="req-is-error",
+        )
+
+    assert _get_response(result).result["isError"] is True
+    (tx,) = events
+    (span,) = tx["spans"]
+    assert span["data"][SPANDATA.ERROR_TYPE] == "tool_error"
+    assert "status" not in span
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize("span_streaming", [True, False])
 @pytest.mark.parametrize(