fix: TypedDict total=False output breaks validation (#816)

When a tool’s output TypedDict uses total=False, MCP clients reject the
response with:
```
MCP error -32602: Structured content does not match the tool's output schema
```
Note that the bug also exists for the Engine transport
(/worker/tools/execute), but since the engine doesn't validate the
output schema, the bug never surfaced. This PR addresses the problem
holistically (MCP and Engine) in preparation for a future where the
Engine transport validates output schemas.

Two bugs combined to cause this:
1. Schema: The outputSchema had no required array and declared all
fields as strict types (e.g. "type": "string"), making every field look
mandatory and non-null.
2. Serialization: model_dump() on TypedDict-derived Pydantic models
emitted None for absent optional fields. A tool returning {"name":
"hello"} produced {"name": "hello", "optional_field": null} which is a
value the schema forbids.


<!-- CURSOR_SUMMARY -->
---

> [!NOTE]
> **Medium Risk**
> Adjusts core schema generation and MCP JSON Schema conversion for
TypedDicts, affecting how tool input/output contracts are emitted and
validated across clients; mistakes could break compatibility or
validation behavior.
> 
> **Overview**
> Fixes MCP/engine validation failures for `TypedDict(total=False)`
outputs by ensuring absent optional keys are **omitted from serialized
output** and that emitted schemas correctly describe **required vs
optional** keys.
> 
> `arcade-core` now tracks `required_keys`/`inner_required_keys` and
per-field `nullable` in `ValueSchema`, derives required sets from
TypedDict `__required_keys__`, and unwraps `Optional[T]` to support
optional nested TypedDicts; TypedDict-derived Pydantic models now
`model_dump(exclude_unset=True)` to avoid leaking missing fields as
`null`.
> 
> `arcade-mcp-server` JSON Schema conversion now emits `required` arrays
(including for arrays of objects), supports `nullable` by generating
`type: [<type>, "null"]` (and `enum` including `None`), and treats
nullable top-level objects as valid unwrapped output schemas. Adds
focused unit/end-to-end tests plus an expanded example server
demonstrating total-false, mixed required/optional, nullable, and
optional-nested TypedDict outputs, and bumps package
versions/dependencies accordingly.
> 
> <sup>Reviewed by [Cursor Bugbot](https://cursor.com/bugbot) for commit
53fe8365f613053599130520b75f30b614b465ca. Bugbot is set up for automated
code reviews on this repo. Configure
[here](https://www.cursor.com/dashboard/bugbot).</sup>
<!-- /CURSOR_SUMMARY -->

examples/mcp_servers/tools_with_output_schema/pyproject.toml

@@ -4,7 +4,7 @@ version = "0.1.0"
 description = "MCP Server created with Arcade.dev"
 requires-python = ">=3.10"
 dependencies = [
-    "arcade-mcp-server>=1.19.0,<2.0.0",
+    "arcade-mcp-server>=1.19.1,<2.0.0",
 ]
 
 [project.optional-dependencies]

examples/mcp_servers/tools_with_output_schema/src/tools_with_output_schema/server.py

@@ -7,17 +7,21 @@ fully-expanded JSON Schema output schemas, so MCP clients can validate and
 display tool results without guessing the shape of the data.
 
 Tools in this server progress from simple to complex:
-  - calculate_statistics  — flat TypedDict (all scalar fields)
-  - analyze_text          — TypedDict with a list field
-  - get_calendar_info     — TypedDict with a nested TypedDict field
-  - parse_url             — TypedDict with two levels of nesting
+  - calculate_statistics     — flat TypedDict (all scalar fields)
+  - analyze_text             — TypedDict with a list field
+  - get_calendar_info        — TypedDict with a nested TypedDict field
+  - parse_url                — TypedDict with two levels of nesting
+  - search_users             — TypedDict with total=False (all optional fields)
+  - get_user_profile         — mixed required/optional fields via inheritance
+  - lookup_record            — nullable fields (str | None)
+  - get_team_info            — optional nested TypedDict (Optional[TypedDict])
 """
 
 import sys
 from collections import Counter
 from datetime import datetime
 from statistics import mean, median
-from typing import Annotated
+from typing import Annotated, Optional
 from urllib.parse import urlparse
 
 from arcade_mcp_server import MCPApp
@@ -88,6 +92,69 @@ class ParsedUrl(TypedDict):
     domain: str
 
 
+class SearchResult(TypedDict, total=False):
+    """Search result where every field is optional (total=False).
+
+    When a field is absent from the returned dict it must NOT appear in the
+    serialized output.  Before the total=False fix, absent fields leaked as
+    explicit ``null`` values, which violated the output schema.
+    """
+
+    username: str
+    email: str
+    display_name: str
+    avatar_url: str
+
+
+class _UserBase(TypedDict):
+    """Required fields that every user profile must include."""
+
+    user_id: int
+    username: str
+
+
+class UserProfile(_UserBase, total=False):
+    """Mixed required / optional fields via TypedDict inheritance.
+
+    ``user_id`` and ``username`` are required (from _UserBase);
+    ``bio`` and ``website`` are optional (total=False on this class).
+    The output schema's ``required`` array must list only the required keys.
+    """
+
+    bio: str
+    website: str
+
+
+class LookupResult(TypedDict):
+    """Demonstrates nullable fields (``str | None``).
+
+    A nullable field can hold either a real value or ``null``, and the
+    output schema must advertise the type as ``["string", "null"]``.
+    """
+
+    key: str
+    value: str | None
+    error_message: str | None
+
+
+class TeamMember(TypedDict, total=False):
+    """A team member with all-optional fields."""
+
+    name: str
+    role: str
+
+
+class TeamInfo(TypedDict):
+    """Demonstrates an optional nested TypedDict (``Optional[TeamMember]``).
+
+    The ``lead`` field is required-but-nullable: it must be present in the
+    output, but its value may be ``null`` when no lead is assigned.
+    """
+
+    team_name: str
+    lead: Optional[TeamMember]
+
+
 # ---------------------------------------------------------------------------
 # Tools
 # ---------------------------------------------------------------------------
@@ -176,6 +243,76 @@ def parse_url(
     )
 
 
+@app.tool
+def search_users(
+    query: Annotated[str, "Search query to match against usernames"],
+) -> Annotated[SearchResult, "Matching user (only populated fields are returned)"]:
+    """Search for a user and return only the fields that matched.
+
+    Demonstrates total=False: absent fields are omitted from the output
+    rather than serialized as null.
+    """
+    # Simulate a search that only finds partial information
+    result = SearchResult(username=query.lower())
+    if "@" in query:
+        result["email"] = query.lower()
+    return result
+
+
+@app.tool
+def get_user_profile(
+    username: Annotated[str, "The username to look up"],
+) -> Annotated[UserProfile, "User profile with required and optional fields"]:
+    """Look up a user profile.
+
+    Demonstrates mixed required/optional fields: user_id and username are
+    always present; bio and website may be absent.
+    """
+    profile = UserProfile(user_id=42, username=username)
+    if username == "admin":
+        profile["bio"] = "Site administrator"
+        profile["website"] = "https://example.com"
+    # For any other user, bio and website are intentionally absent
+    return profile
+
+
+@app.tool
+def lookup_record(
+    key: Annotated[str, "The key to look up"],
+) -> Annotated[LookupResult, "The lookup result with nullable value and error fields"]:
+    """Look up a record by key.
+
+    Demonstrates nullable fields: value and error_message are typed as
+    str | None, so the output schema advertises ["string", "null"].
+    """
+    records = {"color": "blue", "size": "large"}
+    value = records.get(key)
+    return LookupResult(
+        key=key,
+        value=value,
+        error_message=None if value else f"No record found for key: {key}",
+    )
+
+
+@app.tool
+def get_team_info(
+    team_name: Annotated[str, "The team name to look up"],
+) -> Annotated[TeamInfo, "Team info with an optional nested TypedDict for the lead"]:
+    """Get team information including the team lead.
+
+    Demonstrates Optional[TypedDict]: the lead field is required-but-nullable.
+    When a lead exists, absent total=False fields inside the nested TeamMember
+    are properly omitted (not serialized as null).
+    """
+    if team_name == "backend":
+        return TeamInfo(
+            team_name=team_name,
+            lead=TeamMember(name="Alice"),  # role intentionally absent
+        )
+    # Team with no lead assigned
+    return TeamInfo(team_name=team_name, lead=None)
+
+
 # ---------------------------------------------------------------------------
 # Entry point
 # ---------------------------------------------------------------------------

libs/arcade-core/arcade_core/catalog.py

@@ -14,6 +14,7 @@ from typing import (
     Annotated,
     Any,
     Literal,
+    Optional,
     Union,
     cast,
     get_args,
@@ -102,6 +103,9 @@ class WireTypeInfo:
     properties: dict[str, "WireTypeInfo"] | None = None
     inner_properties: dict[str, "WireTypeInfo"] | None = None
     description: str | None = None
+    required_keys: list[str] | None = None
+    inner_required_keys: list[str] | None = None
+    nullable: bool | None = None
 
 
 class ToolMeta(BaseModel):
@@ -762,6 +766,7 @@ def get_wire_type_info(_type: type) -> WireTypeInfo:
     # If so, get the inner (enclosed) type
     is_list = get_origin(_type) is list
     inner_properties = None
+    inner_required_keys = None
 
     if is_list:
         inner_type = get_args(_type)[0]
@@ -773,9 +778,11 @@ def get_wire_type_info(_type: type) -> WireTypeInfo:
         # If inner type has properties (it's a complex object), propagate them
         if inner_info.properties:
             inner_properties = inner_info.properties
+            inner_required_keys = inner_info.required_keys
         # If inner type is array (nested arrays), propagate inner_properties
         elif inner_info.inner_properties:
             inner_properties = inner_info.inner_properties
+            inner_required_keys = inner_info.inner_required_keys
     else:
         inner_wire_type = None
 
@@ -803,8 +810,9 @@ def get_wire_type_info(_type: type) -> WireTypeInfo:
 
     # Extract properties for complex types
     properties = None
+    required_keys = None
     if wire_type == "json" and not is_list:
-        properties = extract_properties(type_to_check)
+        properties, required_keys = extract_properties(type_to_check)
 
     return WireTypeInfo(
         wire_type,
@@ -812,6 +820,8 @@ def get_wire_type_info(_type: type) -> WireTypeInfo:
         enum_values if is_enum else None,
         properties,
         inner_properties,
+        required_keys=required_keys,
+        inner_required_keys=inner_required_keys,
     )
 
 
@@ -844,9 +854,14 @@ def _extract_typeddict_field_descriptions(typeddict_class: type) -> dict[str, st
     return descriptions
 
 
-def extract_properties(type_to_check: type) -> dict[str, WireTypeInfo] | None:
+def extract_properties(
+    type_to_check: type,
+) -> tuple[dict[str, WireTypeInfo] | None, list[str] | None]:
     """
     Extract properties from TypedDict, Pydantic models, or other structured types.
+
+    Returns (properties, required_keys). required_keys is a sorted list of required
+    property names for TypedDict types, or None for other types.
     """
     properties = {}
 
@@ -868,6 +883,8 @@ def extract_properties(type_to_check: type) -> dict[str, WireTypeInfo] | None:
             wire_info = get_wire_type_info(field_type)
             properties[field_name] = wire_info
 
+        return (properties or None, None)
+
     # Handle TypedDict
     elif is_typeddict(type_to_check):
         # Get type hints for the TypedDict
@@ -878,10 +895,13 @@ def extract_properties(type_to_check: type) -> dict[str, WireTypeInfo] | None:
 
         for field_name, field_type in type_hints.items():
             # Handle Optional types (Union[T, None])
-            if is_strict_optional(field_type):
+            is_nullable = is_strict_optional(field_type)
+            if is_nullable:
                 # Extract the non-None type from Optional
                 field_type = next(arg for arg in get_args(field_type) if arg is not type(None))
             wire_info = get_wire_type_info(field_type)
+            if is_nullable:
+                wire_info.nullable = True
 
             # Add description if available
             if field_name in field_descriptions:
@@ -889,12 +909,16 @@ def extract_properties(type_to_check: type) -> dict[str, WireTypeInfo] | None:
 
             properties[field_name] = wire_info
 
+        req = sorted(getattr(type_to_check, "__required_keys__", frozenset()))
+        required_keys = req or None  # normalize empty → None
+        return (properties or None, required_keys)
+
     # Handle regular dict with type annotations (e.g., dict[str, Any])
     elif get_origin(type_to_check) is dict:
         # For generic dicts, we can't extract specific properties
-        return None
+        return (None, None)
 
-    return properties if properties else None
+    return (properties or None, None)
 
 
 def wire_type_info_to_value_schema(wire_info: WireTypeInfo) -> ValueSchema:
@@ -924,6 +948,9 @@ def wire_type_info_to_value_schema(wire_info: WireTypeInfo) -> ValueSchema:
         properties=properties,
         inner_properties=inner_properties,
         description=wire_info.description,
+        required_keys=wire_info.required_keys,
+        inner_required_keys=wire_info.inner_required_keys,
+        nullable=wire_info.nullable,
     )
 
 
@@ -1171,6 +1198,18 @@ def determine_output_model(func: Callable) -> type[BaseModel]:
         )
 
 
+class _TypedDictBaseModel(BaseModel):
+    """Base for Pydantic models derived from TypedDict.
+
+    Defaults model_dump() to exclude_unset=True so that absent optional
+    fields (total=False) don't appear as None in serialized output.
+    """
+
+    def model_dump(self, **kwargs: Any) -> dict[str, Any]:
+        kwargs.setdefault("exclude_unset", True)
+        return super().model_dump(**kwargs)
+
+
 def create_model_from_typeddict(typeddict_class: type, model_name: str) -> type[BaseModel]:
     """
     Create a Pydantic model from a TypedDict class.
@@ -1185,13 +1224,22 @@ def create_model_from_typeddict(typeddict_class: type, model_name: str) -> type[
         # Check if field is required
         is_required = field_name in getattr(typeddict_class, "__required_keys__", set())
 
-        # Handle nested TypedDict
-        if is_typeddict(field_type):
-            nested_model = create_model_from_typeddict(field_type, f"{model_name}_{field_name}")
+        # Unwrap Optional[T] (i.e. T | None) so we can detect nested TypedDicts
+        is_optional_type = is_strict_optional(field_type)
+        inner_type = field_type
+        if is_optional_type:
+            inner_type = next(arg for arg in get_args(field_type) if arg is not type(None))
+
+        # Handle nested TypedDict (works for both T and Optional[T] after unwrapping)
+        if is_typeddict(inner_type):
+            nested_model = create_model_from_typeddict(inner_type, f"{model_name}_{field_name}")
             if is_required:
-                field_definitions[field_name] = (nested_model, Field())
+                if is_optional_type:
+                    field_definitions[field_name] = (Optional[nested_model], Field())
+                else:
+                    field_definitions[field_name] = (nested_model, Field())
             else:
-                field_definitions[field_name] = (nested_model, Field(default=None))
+                field_definitions[field_name] = (Optional[nested_model], Field(default=None))
         else:
             if is_required:
                 field_definitions[field_name] = (field_type, Field())
@@ -1199,7 +1247,7 @@ def create_model_from_typeddict(typeddict_class: type, model_name: str) -> type[
                 field_definitions[field_name] = (field_type, Field(default=None))
 
     # Create and return the Pydantic model
-    return create_model(model_name, **field_definitions)
+    return create_model(model_name, __base__=_TypedDictBaseModel, **field_definitions)
 
 
 def to_tool_secret_requirements(

libs/arcade-core/arcade_core/schema.py

@@ -120,6 +120,15 @@ class ValueSchema(BaseModel):
     description: str | None = None
     """Optional description of the value."""
 
+    required_keys: list[str] | None = None
+    """For object types, the sorted list of required property names."""
+
+    inner_required_keys: list[str] | None = None
+    """For array types with object items, sorted required property names for each item."""
+
+    nullable: bool | None = None
+    """Whether this value can be null (from Optional/T|None annotation). None means not tracked."""
+
 
 class InputParameter(BaseModel):
     """A parameter that can be passed to a tool."""

libs/arcade-core/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "arcade-core"
-version = "4.6.0"
+version = "4.6.1"
 description = "Arcade Core - Core library for Arcade platform"
 readme = "README.md"
 license = { text = "MIT" }

libs/arcade-mcp-server/arcade_mcp_server/convert.py

@@ -219,7 +219,9 @@ def _build_value_schema_json(value_schema: Any) -> dict[str, Any]:
     inner_schema = _value_schema_to_json_schema(value_schema)
 
     # Object return types are already top-level objects, emit directly.
-    if inner_schema.get("type") == "object":
+    # Check for both "object" and ["object", "null"] (nullable top-level TypedDict).
+    schema_type = inner_schema.get("type")
+    if schema_type == "object" or (isinstance(schema_type, list) and "object" in schema_type):
         return inner_schema
 
     # Primitives/arrays must be wrapped so outputSchema.type is "object" per MCP spec.
@@ -231,6 +233,20 @@ def _build_value_schema_json(value_schema: Any) -> dict[str, Any]:
     }
 
 
+def _apply_nullable(schema: dict[str, Any], value_schema: Any) -> dict[str, Any]:
+    """If value_schema.nullable, add null to type and enum."""
+    if not getattr(value_schema, "nullable", False):
+        return schema
+    base_type = schema.get("type")
+    if isinstance(base_type, str):
+        schema["type"] = [base_type, "null"]
+    elif isinstance(base_type, list) and "null" not in base_type:
+        schema["type"] = [*base_type, "null"]
+    if "enum" in schema and None not in schema["enum"]:
+        schema["enum"] = [*schema["enum"], None]
+    return schema
+
+
 def _value_schema_to_json_schema(value_schema: Any) -> dict[str, Any]:
     """Convert a ValueSchema to a JSON Schema dict without top-level object wrapping.
 
@@ -248,7 +264,9 @@ def _value_schema_to_json_schema(value_schema: Any) -> dict[str, Any]:
                 schema["properties"][prop_name] = _value_schema_to_json_schema(prop_schema)
                 if getattr(prop_schema, "description", None):
                     schema["properties"][prop_name]["description"] = prop_schema.description
-        return schema
+        if getattr(value_schema, "required_keys", None):
+            schema["required"] = list(value_schema.required_keys)
+        return _apply_nullable(schema, value_schema)
 
     schema = {"type": _map_type_to_json_schema_type(val_type)}
     if getattr(value_schema, "enum", None):
@@ -262,5 +280,7 @@ def _value_schema_to_json_schema(value_schema: Any) -> dict[str, Any]:
                 items_schema["properties"][prop_name] = _value_schema_to_json_schema(prop_schema)
                 if getattr(prop_schema, "description", None):
                     items_schema["properties"][prop_name]["description"] = prop_schema.description
+            if getattr(value_schema, "inner_required_keys", None):
+                items_schema["required"] = list(value_schema.inner_required_keys)
         schema["items"] = items_schema
-    return schema
+    return _apply_nullable(schema, value_schema)

libs/arcade-mcp-server/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "arcade-mcp-server"
-version = "1.19.0"
+version = "1.19.1"
 description = "Model Context Protocol (MCP) server framework for Arcade.dev"
 readme = "README.md"
 authors = [{ name = "Arcade.dev" }]
@@ -21,7 +21,7 @@ classifiers = [
 ]
 requires-python = ">=3.10"
 dependencies = [
-    "arcade-core>=4.4.2,<5.0.0",
+    "arcade-core>=4.6.1,<5.0.0",
     "arcade-serve>=3.2.0,<4.0.0",
     "arcade-tdk>=3.6.0,<4.0.0",
     "arcadepy>=1.5.0",

libs/tests/arcade_mcp_server/test_convert.py

@@ -5,7 +5,7 @@ import json
 from typing import Annotated
 
 import pytest
-from arcade_core.catalog import MaterializedTool, ToolMeta, create_func_models
+from arcade_core.catalog import MaterializedTool, ToolCatalog, ToolMeta, create_func_models
 from arcade_core.schema import (
     InputParameter,
     ToolDefinition,
@@ -660,3 +660,252 @@ class TestConvertContentToStructuredContent:
 
         result = convert_content_to_structured_content(Custom())
         assert result == {"result": "custom-str"}
+
+
+class TestOutputSchemaOptionalTypedDictFields:
+    """Test that outputSchema correctly represents optional TypedDict fields.
+
+    Reproduces: When a TypedDict uses total=False, extract_properties() treats
+    every field identically — the outputSchema has no 'required' array, and field
+    types never include 'null'. Combined with model_dump() emitting None for absent
+    fields, the MCP client rejects the response because null doesn't match "string".
+    """
+
+    def _make_tool_and_mcp_tool(self, return_type, annotation_desc="result"):
+        """Helper: register a tool returning `return_type` and get the MCP tool."""
+        from typing_extensions import TypedDict
+
+        @tool
+        def f() -> Annotated[return_type, annotation_desc]:
+            """Test tool."""
+            return {}
+
+        tool_def = ToolCatalog().create_tool_definition(f, toolkit_name="test", toolkit_version="1.0")
+        input_model, output_model = create_func_models(f)
+        meta = ToolMeta(module=f.__module__, toolkit="test")
+        mat_tool = MaterializedTool(
+            tool=f,
+            definition=tool_def,
+            meta=meta,
+            input_model=input_model,
+            output_model=output_model,
+        )
+        return create_mcp_tool(mat_tool)
+
+    def test_total_false_typeddict_schema_allows_absent_fields(self):
+        """The outputSchema for a total=False TypedDict must not require any field.
+
+        JSON Schema: if "required" is absent, all properties are optional — that's fine.
+        But the schema must also allow absent fields to validate. Currently the schema
+        does not emit a "required" array, which accidentally makes all fields optional
+        in JSON Schema terms. However, model_dump() reintroduces None values for the
+        absent fields, and "null" is not valid for "type": "string". The schema must
+        either: (a) include "null" in the type, or (b) the serializer must omit Nones.
+        This test validates the schema side: if a field CAN be null in structuredContent,
+        the schema must accept null.
+        """
+        from typing_extensions import TypedDict
+
+        class AllOptional(TypedDict, total=False):
+            name: str
+            count: int
+
+        mcp_tool = self._make_tool_and_mcp_tool(AllOptional)
+        schema = mcp_tool.outputSchema
+
+        assert schema is not None
+        assert schema["type"] == "object"
+        # The schema must not list any field as required since all are total=False
+        required = schema.get("required", [])
+        assert "name" not in required
+        assert "count" not in required
+
+    def test_mixed_required_optional_schema_marks_required_fields(self):
+        """A TypedDict with both required and optional fields must have a 'required' array.
+
+        Required fields (from the base total=True class) must appear in the
+        schema's 'required' array. Optional fields (from total=False) must not.
+        """
+        from typing_extensions import TypedDict
+
+        class _Base(TypedDict):
+            id: int
+
+        class MixedDict(_Base, total=False):
+            label: str
+
+        mcp_tool = self._make_tool_and_mcp_tool(MixedDict)
+        schema = mcp_tool.outputSchema
+
+        assert schema is not None
+        assert schema["type"] == "object"
+        # "id" is required, "label" is optional
+        required = schema.get("required", [])
+        assert "id" in required, (
+            "Required field 'id' must appear in outputSchema.required "
+            f"but got required={required}"
+        )
+        assert "label" not in required
+
+    def test_structuredcontent_validates_against_output_schema(self):
+        """End-to-end: structuredContent for absent optional fields must match outputSchema.
+
+        Simulates the full pipeline: Pydantic model_dump() round-trip then
+        structuredContent conversion. When a tool omits an optional field,
+        model_dump() reintroduces it as None. The outputSchema says "type": "string",
+        so the MCP client rejects the null value.
+        """
+        from arcade_core.catalog import create_model_from_typeddict
+        from typing_extensions import TypedDict
+
+        class ResponseDict(TypedDict, total=False):
+            name: str
+            optional_detail: str
+
+        # 1. Build outputSchema
+        mcp_tool = self._make_tool_and_mcp_tool(ResponseDict)
+        schema = mcp_tool.outputSchema
+
+        # 2. Simulate the Pydantic round-trip that output.py performs:
+        #    create_model_from_typeddict -> instantiate -> model_dump()
+        pydantic_model = create_model_from_typeddict(ResponseDict, "ResponseDict")
+        instance = pydantic_model(**{"name": "hello"})  # optional_detail absent
+        dumped = instance.model_dump()
+
+        # 3. Convert to structuredContent (what server.py does)
+        structured = convert_content_to_structured_content(dumped)
+
+        # The structured content must validate against the schema.
+        # No field in structuredContent should have a value (like null)
+        # that the schema's type declaration doesn't allow.
+        assert structured is not None
+        for field_name, field_schema in schema.get("properties", {}).items():
+            if field_name in structured:
+                value = structured[field_name]
+                allowed_type = field_schema.get("type")
+                if value is None:
+                    # null must be allowed by the schema
+                    if isinstance(allowed_type, list):
+                        assert "null" in allowed_type, (
+                            f"Field '{field_name}' is null in structuredContent but schema "
+                            f"type {allowed_type} does not include 'null'"
+                        )
+                    else:
+                        assert allowed_type == "null" or allowed_type is None, (
+                            f"Field '{field_name}' is null in structuredContent but schema "
+                            f"type is '{allowed_type}', not 'null'"
+                        )
+
+    def test_list_of_typeddict_items_have_required(self):
+        """list[TypedDict] with total=True produces items.required in MCP outputSchema."""
+        from typing_extensions import TypedDict
+
+        class ItemDict(TypedDict):
+            name: str
+            value: int
+
+        mcp_tool = self._make_tool_and_mcp_tool(list[ItemDict])
+        schema = mcp_tool.outputSchema
+
+        assert schema is not None
+        # list output gets wrapped: {type: object, properties: {result: {type: array, ...}}}
+        result_prop = schema["properties"]["result"]
+        assert result_prop["type"] == "array"
+        items_schema = result_prop["items"]
+        assert items_schema["type"] == "object"
+        assert sorted(items_schema["required"]) == ["name", "value"]
+
+    def test_nullable_field_allows_null_in_schema(self):
+        """str | None field produces 'type': ['string', 'null'] in outputSchema."""
+        from typing_extensions import TypedDict
+
+        class NullableDict(TypedDict):
+            label: str
+            note: str | None
+
+        mcp_tool = self._make_tool_and_mcp_tool(NullableDict)
+        schema = mcp_tool.outputSchema
+
+        assert schema is not None
+        props = schema["properties"]
+        assert props["label"]["type"] == "string"
+        assert props["note"]["type"] == ["string", "null"]
+
+    def test_nullable_enum_field_allows_null(self):
+        """Literal['a', 'b'] | None field produces type=['string', 'null'], enum=['a', 'b', None]."""
+        from typing import Literal
+
+        from typing_extensions import TypedDict
+
+        class EnumNullableDict(TypedDict):
+            status: Literal["a", "b"] | None
+
+        mcp_tool = self._make_tool_and_mcp_tool(EnumNullableDict)
+        schema = mcp_tool.outputSchema
+
+        assert schema is not None
+        status_schema = schema["properties"]["status"]
+        assert status_schema["type"] == ["string", "null"]
+        assert status_schema["enum"] == ["a", "b", None]
+
+    def test_input_schema_typeddict_required_keys(self):
+        """TypedDict used as input parameter gets required array in inputSchema."""
+        from typing_extensions import TypedDict
+
+        class ConfigDict(TypedDict):
+            host: str
+            port: int
+
+        @tool
+        def f(config: Annotated[ConfigDict, "The config"]) -> str:
+            """Test tool."""
+            return ""
+
+        tool_def = ToolCatalog().create_tool_definition(
+            f, toolkit_name="test", toolkit_version="1.0"
+        )
+        input_model, output_model = create_func_models(f)
+        meta = ToolMeta(module=f.__module__, toolkit="test")
+        mat_tool = MaterializedTool(
+            tool=f,
+            definition=tool_def,
+            meta=meta,
+            input_model=input_model,
+            output_model=output_model,
+        )
+        mcp_tool = create_mcp_tool(mat_tool)
+        config_schema = mcp_tool.inputSchema["properties"]["config"]
+
+        assert config_schema["type"] == "object"
+        assert sorted(config_schema["required"]) == ["host", "port"]
+
+    def test_input_schema_typeddict_nullable_field(self):
+        """TypedDict input parameter with str | None field gets type=['string', 'null']."""
+        from typing_extensions import TypedDict
+
+        class InputDict(TypedDict):
+            name: str
+            tag: str | None
+
+        @tool
+        def f(data: Annotated[InputDict, "The data"]) -> str:
+            """Test tool."""
+            return ""
+
+        tool_def = ToolCatalog().create_tool_definition(
+            f, toolkit_name="test", toolkit_version="1.0"
+        )
+        input_model, output_model = create_func_models(f)
+        meta = ToolMeta(module=f.__module__, toolkit="test")
+        mat_tool = MaterializedTool(
+            tool=f,
+            definition=tool_def,
+            meta=meta,
+            input_model=input_model,
+            output_model=output_model,
+        )
+        mcp_tool = create_mcp_tool(mat_tool)
+        data_schema = mcp_tool.inputSchema["properties"]["data"]
+
+        assert data_schema["properties"]["name"]["type"] == "string"
+        assert data_schema["properties"]["tag"]["type"] == ["string", "null"]

libs/tests/core/test_typeddict_output_execution.py

@@ -86,6 +86,32 @@ def returns_dict_list() -> Annotated[list[dict], "Returns list of dicts"]:
     ]
 
 
+class MixedRequiredDict(TypedDict):
+    """Base TypedDict with required fields."""
+
+    name: str
+
+
+class MixedOptionalDict(MixedRequiredDict, total=False):
+    """TypedDict with both required and optional fields."""
+
+    optional_field: str
+
+
+@tool
+def returns_partial_optional_typeddict() -> Annotated[
+    OptionalFieldsDict, "Returns partial optional TypedDict"
+]:
+    """Tool that returns a TypedDict with some fields omitted."""
+    return {"required_field": "hello"}  # optional_field intentionally omitted
+
+
+@tool
+def returns_mixed_typeddict() -> Annotated[MixedOptionalDict, "Returns mixed TypedDict"]:
+    """Tool that returns a TypedDict with required + optional fields, omitting optional."""
+    return {"name": "hello"}  # optional_field intentionally omitted
+
+
 class TestTypeDictOutputExecution:
     """Test TypedDict outputs through the full execution pipeline."""
 
@@ -248,3 +274,137 @@ class TestTypeDictOutputExecution:
             {"name": "string", "data": "test"},
             {"name": "typed", "value": 99},  # TypedDict becomes regular dict at runtime
         ]
+
+    @pytest.mark.asyncio
+    async def test_total_false_typeddict_omits_absent_fields(self, catalog, context):
+        """When a total=False TypedDict omits a field, it must not appear as None in output.
+
+        Reproduces: model_dump() without exclude_none reintroduces None for absent
+        optional fields, which then fails JSON Schema validation against the outputSchema
+        (the schema declares "type": "string", not ["string", "null"]).
+        """
+        definition = catalog.create_tool_definition(
+            returns_partial_optional_typeddict, toolkit_name="test", toolkit_version="1.0.0"
+        )
+        input_model, output_model = create_func_models(returns_partial_optional_typeddict)
+
+        result = await ToolExecutor.run(
+            func=returns_partial_optional_typeddict,
+            definition=definition,
+            input_model=input_model,
+            output_model=output_model,
+            context=context,
+        )
+
+        assert result.error is None
+        # The absent optional field must NOT be present with a None value
+        assert result.value == {"required_field": "hello"}
+        assert "optional_field" not in result.value
+
+    @pytest.mark.asyncio
+    async def test_mixed_required_optional_typeddict_omits_absent_fields(self, catalog, context):
+        """A TypedDict inheriting required fields + total=False optional fields.
+
+        When the tool omits the optional field, the output must contain only the
+        required field. None values for absent optional fields corrupt the MCP
+        structuredContent because they don't match the schema type.
+        """
+        definition = catalog.create_tool_definition(
+            returns_mixed_typeddict, toolkit_name="test", toolkit_version="1.0.0"
+        )
+        input_model, output_model = create_func_models(returns_mixed_typeddict)
+
+        result = await ToolExecutor.run(
+            func=returns_mixed_typeddict,
+            definition=definition,
+            input_model=input_model,
+            output_model=output_model,
+            context=context,
+        )
+
+        assert result.error is None
+        assert result.value == {"name": "hello"}
+        assert "optional_field" not in result.value
+
+    @pytest.mark.asyncio
+    async def test_exclude_unset_omits_absent_but_keeps_explicit_none(self, catalog, context):
+        """TypedDict with total=False absent field AND total=True str|None field set to None.
+
+        The absent field must be excluded, but the explicitly-set None must be kept.
+        This validates exclude_unset (not exclude_none) semantics.
+        """
+
+        class _Required(TypedDict):
+            name: str
+            note: str | None  # total=True, explicitly nullable
+
+        class _WithOptional(_Required, total=False):
+            tag: str  # total=False, will be absent
+
+        @tool
+        def returns_mixed_nullable() -> Annotated[_WithOptional, "Mixed nullable output"]:
+            """Tool with both nullable required and absent optional fields."""
+            return {"name": "hello", "note": None}  # tag intentionally absent
+
+        definition = catalog.create_tool_definition(
+            returns_mixed_nullable, toolkit_name="test", toolkit_version="1.0.0"
+        )
+        input_model, output_model = create_func_models(returns_mixed_nullable)
+
+        result = await ToolExecutor.run(
+            func=returns_mixed_nullable,
+            definition=definition,
+            input_model=input_model,
+            output_model=output_model,
+            context=context,
+        )
+
+        assert result.error is None
+        # Explicit None must be kept (exclude_unset preserves it)
+        assert result.value == {"name": "hello", "note": None}
+        # Absent total=False field must NOT appear
+        assert "tag" not in result.value
+
+    @pytest.mark.asyncio
+    async def test_optional_nested_typeddict_omits_absent_fields(self, catalog, context):
+        """A TypedDict field typed as Optional[NestedTypedDict] where the nested
+        TypedDict has total=False.
+
+        When the nested TypedDict omits optional fields, those fields must NOT
+        appear as None in the serialized output. This validates that
+        create_model_from_typeddict unwraps Optional before checking is_typeddict,
+        so the nested TypedDict gets the _TypedDictBaseModel (exclude_unset) treatment.
+        """
+
+        class _InnerPartial(TypedDict, total=False):
+            label: str
+            count: int
+
+        class _OuterWithOptionalNested(TypedDict):
+            id: int
+            nested: Optional[_InnerPartial]
+
+        @tool
+        def returns_optional_nested_partial() -> Annotated[
+            _OuterWithOptionalNested, "Outer with optional nested partial TypedDict"
+        ]:
+            """Tool returning an Optional nested TypedDict with absent fields."""
+            return {"id": 1, "nested": {"label": "hello"}}  # count intentionally absent
+
+        definition = catalog.create_tool_definition(
+            returns_optional_nested_partial, toolkit_name="test", toolkit_version="1.0.0"
+        )
+        input_model, output_model = create_func_models(returns_optional_nested_partial)
+
+        result = await ToolExecutor.run(
+            func=returns_optional_nested_partial,
+            definition=definition,
+            input_model=input_model,
+            output_model=output_model,
+            context=context,
+        )
+
+        assert result.error is None
+        assert result.value == {"id": 1, "nested": {"label": "hello"}}
+        # The absent total=False field in the nested TypedDict must NOT appear
+        assert "count" not in result.value["nested"]

libs/tests/tool/test_create_tool_definition_typeddict.py

@@ -161,6 +161,7 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                             "price": ValueSchema(val_type="integer", enum=None),
                             "stock_quantity": ValueSchema(val_type="integer", enum=None),
                         },
+                        required_keys=["price", "product_name", "stock_quantity"],
                     ),
                     available_modes=["value", "error"],
                     description="The product, price, and quantity",
@@ -181,6 +182,7 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                             "price": ValueSchema(val_type="integer", enum=None),
                             "stock_quantity": ValueSchema(val_type="integer", enum=None),
                         },
+                        inner_required_keys=["price", "product_name", "stock_quantity"],
                     ),
                     available_modes=["value", "error"],
                     description="The product, price, and quantity",
@@ -206,6 +208,7 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                                     "price": ValueSchema(val_type="integer", enum=None),
                                     "stock_quantity": ValueSchema(val_type="integer", enum=None),
                                 },
+                                required_keys=["price", "product_name", "stock_quantity"],
                             ),
                         )
                     ]
@@ -226,6 +229,7 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                             "stock_quantity": ValueSchema(val_type="integer", enum=None),
                             "description": ValueSchema(val_type="string", enum=None, nullable=True),
                         },
+                        required_keys=["description", "price", "product_name", "stock_quantity"],
                     ),
                     available_modes=["value", "error"],
                     description="The product, price, and quantity",
@@ -252,6 +256,7 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                                         val_type="array", inner_val_type="string", enum=None
                                     ),
                                 },
+                                required_keys=["category", "products"],
                             ),
                         )
                     ]
@@ -305,8 +310,10 @@ def func_returns_nested_typedicts() -> Annotated[CustomerDict, "Customer informa
                                     "city": ValueSchema(val_type="string", enum=None),
                                     "zip_code": ValueSchema(val_type="string", enum=None),
                                 },
+                                required_keys=["city", "street", "zip_code"],
                             ),
                         },
+                        required_keys=["address", "email", "name"],
                     ),
                     available_modes=["value", "error"],
                     description="Customer information with address",