diff --git a/superset/mcp_service/auth.py b/superset/mcp_service/auth.py
index 1bb5510a806..02ca11f0cc2 100644
--- a/superset/mcp_service/auth.py
+++ b/superset/mcp_service/auth.py
@@ -277,6 +277,23 @@ def _setup_user_context() -> User | None:
             logger.debug("No Flask app context available for user setup")
             return None
         raise
+    except ValueError as e:
+        # JWT user resolution failed (e.g. SAML subject not in DB).
+        # If middleware already set g.user (request context exists),
+        # use that instead of failing closed.
+        from flask import has_request_context
+
+        if has_request_context() and hasattr(g, "user") and g.user:
+            logger.warning(
+                "JWT user resolution failed (%s), using middleware-provided g.user=%s",
+                e,
+                g.user.username,
+            )
+            # Assign to local so relationship validation below runs
+            # (same as the normal path) to prevent detached instance errors.
+            user = g.user
+        else:
+            raise
 
     # Validate user has necessary relationships loaded
     # (Force access to ensure they're loaded if lazy)
diff --git a/superset/mcp_service/dashboard/tool/add_chart_to_existing_dashboard.py b/superset/mcp_service/dashboard/tool/add_chart_to_existing_dashboard.py
index a32fef3cfea..83967209c05 100644
--- a/superset/mcp_service/dashboard/tool/add_chart_to_existing_dashboard.py
+++ b/superset/mcp_service/dashboard/tool/add_chart_to_existing_dashboard.py
@@ -415,12 +415,8 @@ def add_chart_to_existing_dashboard(
             published=updated_dashboard.published,
             created_on=updated_dashboard.created_on,
             changed_on=updated_dashboard.changed_on,
-            created_by=updated_dashboard.created_by.username
-            if updated_dashboard.created_by
-            else None,
-            changed_by=updated_dashboard.changed_by.username
-            if updated_dashboard.changed_by
-            else None,
+            created_by=updated_dashboard.created_by_name or None,
+            changed_by=updated_dashboard.changed_by_name or None,
             uuid=str(updated_dashboard.uuid) if updated_dashboard.uuid else None,
             url=f"{get_superset_base_url()}/superset/dashboard/{updated_dashboard.id}/",
             chart_count=len(updated_dashboard.slices),
diff --git a/superset/mcp_service/dashboard/tool/generate_dashboard.py b/superset/mcp_service/dashboard/tool/generate_dashboard.py
index 8feee01ab68..5f0ed4ee5cc 100644
--- a/superset/mcp_service/dashboard/tool/generate_dashboard.py
+++ b/superset/mcp_service/dashboard/tool/generate_dashboard.py
@@ -231,10 +231,8 @@ def generate_dashboard(
 
         # Prepare dashboard data and create dashboard
         with event_logger.log_context(action="mcp.generate_dashboard.db_write"):
-            dashboard_data = {
+            dashboard_data: Dict[str, Any] = {
                 "dashboard_title": dashboard_title,
-                "slug": None,  # Let Superset auto-generate slug
-                "css": "",
                 "json_metadata": json.dumps(
                     {
                         "filter_scopes": {},
@@ -265,8 +263,23 @@ def generate_dashboard(
                 dashboard_data["description"] = request.description
 
             # Create the dashboard using Superset's command pattern
-            command = CreateDashboardCommand(dashboard_data)
-            dashboard = command.run()
+            try:
+                command = CreateDashboardCommand(dashboard_data)
+                dashboard = command.run()
+            except Exception as cmd_err:
+                # Surface the root cause from @transaction's error wrapping
+                root_cause = cmd_err.__cause__ or cmd_err
+                logger.error(
+                    "CreateDashboardCommand failed: %s (cause: %s)",
+                    cmd_err,
+                    root_cause,
+                    exc_info=True,
+                )
+                return GenerateDashboardResponse(
+                    dashboard=None,
+                    dashboard_url=None,
+                    error=f"Failed to create dashboard: {root_cause}",
+                )
 
         # Convert to our response format
         from superset.mcp_service.dashboard.schemas import (
@@ -282,8 +295,8 @@ def generate_dashboard(
             published=dashboard.published,
             created_on=dashboard.created_on,
             changed_on=dashboard.changed_on,
-            created_by=dashboard.created_by.username if dashboard.created_by else None,
-            changed_by=dashboard.changed_by.username if dashboard.changed_by else None,
+            created_by=dashboard.created_by_name or None,
+            changed_by=dashboard.changed_by_name or None,
             uuid=str(dashboard.uuid) if dashboard.uuid else None,
             url=f"{get_superset_base_url()}/superset/dashboard/{dashboard.id}/",
             chart_count=len(request.chart_ids),
diff --git a/superset/mcp_service/middleware.py b/superset/mcp_service/middleware.py
index 0fadd648ae8..b0ffc4f5f7c 100644
--- a/superset/mcp_service/middleware.py
+++ b/superset/mcp_service/middleware.py
@@ -18,10 +18,13 @@
 import logging
 import time
 from collections import defaultdict
-from typing import Any, Awaitable, Callable, Dict, Protocol
+from typing import Any, Awaitable, Callable, Dict, Protocol, Sequence
 
+import mcp.types as mt
 from fastmcp.exceptions import ToolError
 from fastmcp.server.middleware import Middleware, MiddlewareContext
+from fastmcp.server.middleware.middleware import CallNext
+from fastmcp.tools.tool import Tool, ToolResult
 from flask import has_app_context
 from pydantic import ValidationError
 from sqlalchemy.exc import OperationalError, TimeoutError
@@ -259,6 +262,62 @@ class PrivateToolMiddleware(Middleware):
         return await call_next(context)
 
 
+class StructuredContentStripperMiddleware(Middleware):
+    """Strip ``outputSchema`` and ``structured_content`` to prevent encoding errors.
+
+    FastMCP 3.x auto-generates ``outputSchema`` in tool definitions
+    (``tools/list``) and ``structuredContent`` in tool call responses
+    (``tools/call``) when the tool has a typed return annotation.
+
+    Some MCP client transports (e.g. Claude.ai's MCP bridge) cannot handle
+    ``structuredContent`` dicts, causing ``TypeError: encoding without a
+    string argument``.  Additionally, if ``outputSchema`` is advertised but
+    ``structuredContent`` is stripped from the response, clients may raise
+    ``Output validation error: outputSchema defined but no structured output
+    returned``.
+
+    This middleware handles both sides:
+    - ``on_list_tools``: removes ``output_schema`` from every tool definition
+    - ``on_call_tool``: removes ``structured_content`` from every tool result
+    """
+
+    async def on_list_tools(
+        self,
+        context: MiddlewareContext[mt.ListToolsRequest],
+        call_next: CallNext[mt.ListToolsRequest, Sequence[Tool]],
+    ) -> Sequence[Tool]:
+        tools = await call_next(context)
+        return [
+            t.model_copy(update={"output_schema": None})
+            if t.output_schema is not None
+            else t
+            for t in tools
+        ]
+
+    async def on_call_tool(
+        self,
+        context: MiddlewareContext[mt.CallToolRequestParams],
+        call_next: Callable[[MiddlewareContext], Awaitable[ToolResult]],
+    ) -> ToolResult:
+        try:
+            result = await call_next(context)
+        except Exception as e:
+            # When exceptions propagate past the middleware chain to the
+            # MCP SDK layer, they become CallToolResult(isError=True).
+            # Some transports (Claude.ai's MCP bridge) cannot encode these
+            # error responses, producing "encoding without a string argument".
+            # Catch ALL exceptions (not just specific types) because any
+            # unhandled exception — including ToolError from
+            # GlobalErrorHandlerMiddleware, ValueError, TypeError, etc. —
+            # will cause encoding failures on the wire.
+            return ToolResult(
+                content=[mt.TextContent(type="text", text=f"Error: {e}")],
+            )
+        if isinstance(result, ToolResult) and result.structured_content is not None:
+            result = ToolResult(content=result.content, meta=result.meta)
+        return result
+
+
 class GlobalErrorHandlerMiddleware(Middleware):
     """
     Global error handler middleware that provides consistent error responses
diff --git a/superset/mcp_service/server.py b/superset/mcp_service/server.py
index 732e8f4622f..aff5cd8b760 100644
--- a/superset/mcp_service/server.py
+++ b/superset/mcp_service/server.py
@@ -39,6 +39,7 @@ from superset.mcp_service.middleware import (
     create_response_size_guard_middleware,
     GlobalErrorHandlerMiddleware,
     LoggingMiddleware,
+    StructuredContentStripperMiddleware,
 )
 from superset.mcp_service.storage import _create_redis_store
 
@@ -363,9 +364,15 @@ def run_server(
         # Add logging middleware (logs all tool calls with duration tracking)
         middleware_list.append(LoggingMiddleware())
 
-        # Add global error handler (outermost – catches all exceptions)
+        # Add global error handler (catches all exceptions, raises ToolError)
         middleware_list.append(GlobalErrorHandlerMiddleware())
 
+        # Strip outputSchema from tool definitions and structuredContent from
+        # tool responses to prevent encoding errors on Claude.ai's MCP bridge.
+        # MUST be outermost so it catches ToolError from GlobalErrorHandler
+        # and converts to plain text before the MCP SDK tries to encode it.
+        middleware_list.append(StructuredContentStripperMiddleware())
+
         mcp_instance = init_fastmcp_server(
             auth=auth_provider,
             middleware=middleware_list or None,
diff --git a/superset/mcp_service/sql_lab/tool/save_sql_query.py b/superset/mcp_service/sql_lab/tool/save_sql_query.py
index 448ce9e1924..65c76e6bb93 100644
--- a/superset/mcp_service/sql_lab/tool/save_sql_query.py
+++ b/superset/mcp_service/sql_lab/tool/save_sql_query.py
@@ -121,10 +121,10 @@ async def save_sql_query(
             id=saved_query.id,
             label=saved_query.label,
             sql=saved_query.sql,
-            database_id=request.database_id,
-            schema_name=request.schema_name,
+            database_id=saved_query.db_id,
+            schema_name=saved_query.schema or None,
             catalog=getattr(saved_query, "catalog", None),
-            description=request.description,
+            description=saved_query.description or None,
             url=saved_query_url,
         )
 
diff --git a/superset/mcp_service/utils/schema_utils.py b/superset/mcp_service/utils/schema_utils.py
index e382ec3dce8..34f99be88d3 100644
--- a/superset/mcp_service/utils/schema_utils.py
+++ b/superset/mcp_service/utils/schema_utils.py
@@ -512,6 +512,7 @@ def _apply_signature_for_fastmcp(
     wrapper: Any,
     original_func: Callable[..., Any],
     request_annotation: Any,
+    request_default: Any = None,
 ) -> None:
     """Apply annotations and signature to wrapper, stripping ctx for FastMCP.
 
@@ -539,7 +540,10 @@ def _apply_signature_for_fastmcp(
         if _is_context_param(param, name, FMContext):
             continue
         if name == "request":
-            new_params.append(param.replace(annotation=request_annotation))
+            replacement = {"annotation": request_annotation}
+            if request_default is not None:
+                replacement["default"] = request_default
+            new_params.append(param.replace(**replacement))
         else:
             new_params.append(param)
     wrapper.__signature__ = orig_sig.replace(parameters=new_params)
diff --git a/superset/sql/execution/executor.py b/superset/sql/execution/executor.py
index e021920023d..698baca29a4 100644
--- a/superset/sql/execution/executor.py
+++ b/superset/sql/execution/executor.py
@@ -839,13 +839,22 @@ class SQLExecutor:
             or app.config.get("CACHE_DEFAULT_TIMEOUT", 300)
         )
 
-        # Serialize statement results for caching
+        # Serialize statement results for caching.
+        # Convert DataFrames to list-of-dicts so the cache backend
+        # does not need to pickle pandas objects (which can fail to
+        # deserialize correctly with some backends or pandas versions).
+        import pandas as pd
+
         cached_data = {
             "statements": [
                 {
                     "original_sql": stmt.original_sql,
                     "executed_sql": stmt.executed_sql,
-                    "data": stmt.data,
+                    "data": (
+                        stmt.data.to_dict(orient="records")
+                        if isinstance(stmt.data, pd.DataFrame)
+                        else stmt.data
+                    ),
                     "row_count": stmt.row_count,
                     "execution_time_ms": stmt.execution_time_ms,
                 }
diff --git a/tests/unit_tests/mcp_service/dashboard/tool/test_dashboard_generation.py b/tests/unit_tests/mcp_service/dashboard/tool/test_dashboard_generation.py
index c990fdde2a5..6aa9b27b688 100644
--- a/tests/unit_tests/mcp_service/dashboard/tool/test_dashboard_generation.py
+++ b/tests/unit_tests/mcp_service/dashboard/tool/test_dashboard_generation.py
@@ -96,6 +96,8 @@ def _mock_dashboard(id: int = 1, title: str = "Test Dashboard") -> Mock:
     dashboard.created_by.username = "test_user"
     dashboard.changed_by = Mock()
     dashboard.changed_by.username = "test_user"
+    dashboard.created_by_name = "test_user"
+    dashboard.changed_by_name = "test_user"
     dashboard.uuid = f"dashboard-uuid-{id}"
     dashboard.slices = []
     dashboard.owners = []
diff --git a/tests/unit_tests/mcp_service/sql_lab/tool/test_save_sql_query.py b/tests/unit_tests/mcp_service/sql_lab/tool/test_save_sql_query.py
index 1d736f18cc9..39443e4ef17 100644
--- a/tests/unit_tests/mcp_service/sql_lab/tool/test_save_sql_query.py
+++ b/tests/unit_tests/mcp_service/sql_lab/tool/test_save_sql_query.py
@@ -226,6 +226,9 @@ class TestSaveSqlQueryToolLogic:
             mock_sq.id = 42
             mock_sq.label = "Revenue Query"
             mock_sq.sql = "SELECT SUM(revenue) FROM sales"
+            mock_sq.db_id = 1
+            mock_sq.schema = ""
+            mock_sq.description = ""
             mock_sq.catalog = None
 
             request = SaveSqlQueryRequest(
@@ -390,6 +393,9 @@ class TestSaveSqlQueryToolLogic:
             mock_sq.id = 10
             mock_sq.label = "Test"
             mock_sq.sql = "SELECT 1"
+            mock_sq.db_id = 1
+            mock_sq.schema = "public"
+            mock_sq.description = ""
             mock_sq.catalog = None
 
             request = SaveSqlQueryRequest(